SlideShare a Scribd company logo
1 of 94
Download to read offline
ย 
An Empirical Assessment of the Role of the
Australian Housing Market in the
Transmission of Monetary Policy
Mollie Annie Urquhart
Economics Honours Dissertation
Submitted in partial fulfilment of the requirements for the
degree of Bachelor of Commerce (Honours)
Supervised by Professor Nicolaas Groenewold
Submission date: 14 November 2016
ย 
	
 ย 
i
Abstract
Since the early 1990s the Reserve Bank of Australia (RBA) has explicitly targeted a medium
term level of inflation of between two and three per cent. As housing costs account for
approximately 22 per cent of the Consumer Price Index basket โ€“ the RBAโ€™s primary measure
of inflation โ€“ and housing investment four per cent of Gross Domestic Product, it is clear the
Australian housing market is intrinsically important to the RBAโ€™s objective; making an
assessment of the role of the housing market in monetary policy transmission an important
exercise.
This dissertation pursues this objective using data for the period 1990Q1 to 2015Q4 and a
structural vector autoregression model to simulate the interaction of house prices, housing
investment and the cash rate in response to a monetary policy shock.
Results indicate that (1) house prices initially react positively to a monetary contraction,
suggesting a form of the โ€˜price puzzleโ€™ is operational in the housing market, (2) the quantity of
housing investment falls in response to a contractionary monetary shock and (3) the cash rate
operates independently of the housing market.
Alternative data sets are used, additional variables are included and sub-periods are created to
determine the robustness of the results; reducing the likelihood of model mis-specification.
The results are remarkably robust.
ย 
	
 ย 
	
 ย 
ii
Acknowledgements 	
 ย 
First and foremost, I would like to thank my supervisor Prof. Nicolaas Groenewold; without
whom this dissertation would not have been possible. Thank you Nic for your guidance and
support throughout this year; I am grateful for how pleasant you have made the experience.
Thank you Elisa Birch, Anu Rammohan and particularly Leandro Magnusson for your
continued guidance. Thank you also to my lecturers Peter Robertson, Rodney Tyers, Michael
McLure and Yanrui Wu.
I would also like to thank my parents Mr David Urquhart and Mrs Wendy Urquhart. The
sacrifices they have made and the support they have provided throughout my entire education
has meant the world. This dissertation is a testament to their belief in my ability.
Thank you Dean van Kwawegen for proofreading this dissertation.
Finally, thank you to the 2016 Economics Honours cohort, it has been a pleasure to write my
dissertation alongside you all.
ย 
	
 ย 
	
 ย 
iii
Declaration
Unless otherwise acknowledged in the text or acknowledgements, the work presented in this
dissertation is my own original work.
This dissertation has 14,684 words, including appendices.
I give permission for the Economic Discipline to use my honours dissertation as an
example of a dissertation which may get distributed to staff and future students.โ€จ
I do not give permission for the Economic Discipline to use my honours dissertation as
an example of a dissertation which may get distributed to staff and future students.
___________________________________ _______________________
Mollie Annie Urquhart Date
14th
November 2016
ย 
	
 ย 
	
 ย 
iv
Table of Contents
Abstract......................................................................................................................................i
Acknowledgements ................................................................................................................. ii
Declaration.............................................................................................................................. iii
Table of Contents....................................................................................................................iv
List of Tables ...........................................................................................................................vi
List of Figures........................................................................................................................ vii
Chapter One โ€“ Introduction ...................................................................................................1
Chapter Two โ€“ Australian Monetary Policy and the Housing Market ..............................4
2.1 Monetary Policy, Then and Now............................................................................................... 4
2.2 Monetary Policy Transmission Channels................................................................................. 5
2.3 The Australian Housing Market ............................................................................................... 7
2.4 Concluding Remarks.................................................................................................................. 9
Chapter Three โ€“ Literature Review.....................................................................................10
3.1 Monetary Policy Literature..................................................................................................... 10
3.2 International Housing Market Literature.............................................................................. 12
3.3 Australian Housing Market Literature.................................................................................. 13
3.4 Asset Price Targeting Literature ............................................................................................ 14
3.5 Price Puzzle Literature ............................................................................................................ 15
3.6 Contribution to the Existing Literature ................................................................................. 16
Chapter Four โ€“ Method and Data .......................................................................................18
4.1 Ordinary Least Squares Method (OLS)................................................................................. 18
4.2 Structural Vector Autoregressive Model (SVAR)................................................................. 19
4.2.1 Identification of Monetary Policy Shock ............................................................................ 20
4.2.2 Impulse Response Functions............................................................................................... 20
4.2.3 Block Exogeneity Wald Test .............................................................................................. 21
4.3 Variables.................................................................................................................................... 21
4.4 Data............................................................................................................................................ 22
4.4.1 Stationarity .......................................................................................................................... 23
4.5 Primary Variable Tests............................................................................................................ 25
ย 
	
 ย 
	
 ย 
v
4.6 Robustness Variables Tests ..................................................................................................... 28
4.6.1 Alternative Data Sets........................................................................................................... 28
4.6.2 Additional Variables ........................................................................................................... 30
Chapter Five - Results ...........................................................................................................32
5.1 Preliminary Analysis................................................................................................................ 32
5.1.1 Ordinary Least Square Regression Analysis....................................................................... 32
5.2 Structural Vector Auto-Regression Model ............................................................................ 34
5.2.1 SVAR (I) ............................................................................................................................. 34
5.2.2 Lag Selection....................................................................................................................... 34
5.2.3 Residual Serial Correlation LM Test .................................................................................. 35
5.2.4 Impulse Response Functions............................................................................................... 36
5.2.5 Formal Test of Cash Rate Exogeneity ................................................................................ 39
5.3 Sensitivity Analysis................................................................................................................... 40
5.3.1 Alternative Variable Representation............................................................................... 41
5.3.2 Additional Variables ....................................................................................................... 43
5.3.3 Sub-Periods ..................................................................................................................... 46
5.3.4 Additional Sensitivity Testing ........................................................................................ 49
5.4 Concluding Remarks................................................................................................................ 52
Chapter Six - Conclusion ......................................................................................................54
References...............................................................................................................................56
Appendix.................................................................................................................................64
ย 
	
 ย 
	
 ย 
vi
List of Tables
Table 4.1 Unit Root Tests.......................................................................................................26
Table 4.2 Breusch-Godfrey Serial Correlation LM Test ........................................................27
Table 4.3 Unit Root Tests.......................................................................................................28
Table 4.4 Breusch-Godfrey Serial Correlation LM Test ........................................................30
Table 4.5 Unit Root Tests.......................................................................................................31
Table 4.6 Breusch-Godfrey Serial Correlation LM Test ........................................................31
Table 5.1 OLS Regression Results ..........................................................................................33
Table 5.2 Optimal VAR Lag Selection Criteria.......................................................................34
Table 5.3 SVAR Residual Serial Correlation LM Test ...........................................................36
Table 5.4 SVAR Block Exogeneity Wald Test .......................................................................40
Table 5.5 SVAR Block Exogeneity Wald Test .......................................................................43
Table 5.6 SVAR Block Exogeneity Wald Test .......................................................................45
Table 5.7 SVAR Block Exogeneity Wald Test .......................................................................48
ย 
	
 ย 
	
 ย 
vii
List of Figures
Figure 1.1 Australian Housing Lending Rates...........................................................................2
Figure 1.2 Housing Prices and Interest Rates............................................................................2
Figure 2.1 Australian Inflation...................................................................................................5
Figure 2.2 Response of the Australian Housing Market to Expansionary Monetary Policy.....9
Figure 4.1 Cash Rate, ln(House Price Index), ln(Dwelling Commencements) Plots..............24
Figure 5.1 Impulse Responses of SVAR(I) over 10 Periods ...................................................38
Figure 5.2 Impulse Responses of FC and CR over 10 Periods................................................42
Figure 5.3 Impulse Response of CR to ACX over 10 Periods.................................................45
Figure 5.4 Impulse Response of HPI to CR over 10 Periods...................................................47
Figure 5.5 Impulse Response of DC to CR over 10 Periods ...................................................48
Figure 5.6 Impulse Response of CR and MR over 10 Periods................................................50
Figure 5.7 Impulse Responses of CR and HPI (Cumulative) over 30 Periods........................51
Figure 5.8 Accumulated Impulse Response of DC (First Differenced) over 10 Periods ........52
ย 
	
 ย 
	
 ย 
1
Chapter One โ€“ Introduction
Since the early 1990s the Reserve Bank of Australia (RBA) has explicitly targeted inflation
levels of between two and three per cent over the medium term. This target is met through the
exercising of monetary policy โ€“ specifically changing the cash rate โ€“ to influence economy
wide interest rates and ultimately manage economic conditions.
Costs of housing investment borne by owner occupiers account for approximately nine per cent
of the Consumer Price Index (CPI) basket โ€“ the primary measure of annual inflation โ€“ while
housing costs in sum compose 22.3 per cent (Reserve Bank of Australia 2014).1
Together
Figure 1.1 and 1.2 detail the strong pass-through of cash rate changes to housing lending rates
and the transmission of these effects to house prices. A strong positive relationship is
observable between the cash rate and mortgage rate, and a strong negative relationship is
evident between the mortgage rate and house prices. As the cash rate holds a strong relationship
with house prices, through the mortgage rate, and house prices significantly contribute to
inflation, through the CPI, it appears that the transmission of monetary policy to house prices
is important to the maintenance of the inflation target. It is surprising, in light of the importance
of Australian housing to CPI inflation measurement, and the strong relationship between the
cash rate, mortgage rate and house prices, that an assessment of the Australian housing
marketโ€™s role in the maintenance of the inflation target has not previously been conducted.
1
The housing component of the Consumer Price Index is calculated by the Australian Bureau of Statistics
(ABS) using costs of rent, new dwellings purchased by owner occupiers, maintenance and repair costs, property
rates and charges, utilities and household products. In aggregate the housing component accounts for 22.3 per
cent of the CPI basket. While rent costs account for 6.71 per cent CPI and new dwelling owner occupier costs
account for 8.67 per cent of the CPI.
ย 
	
 ย 
	
 ย 
2
Our research is interested in the transmission of monetary policy shocks to inflation and within
this transmission system we are specifically concerned with the role of the housing market. We
analyse the effects of changes in the cash rate on house prices and housing investment; which
are, respectively, transmitted to inflation directly, through CPI measurement and indirectly,
through aggregate demand. Observable in Figure 1.1 and 1.2 is the direct transmission of
monetary policy to inflation through house prices. Additionally, an indirect effect upon
inflation, transmitted through housing investment is anticipated to be operational. This is a
consequence of the intrinsic importance of housing investment to the Australian economy,
accounting for approximately four per cent of Gross Domestic Product (GDP) (Australian
Industry Group 2015), while also tending to be the most interest rate sensitive component
(Berger - Thomson & Ellis 2014), and thus a strong driving force of aggregate demand and
ultimately inflation.
The relationship between monetary policy and the housing market is noteworthy not only
because of these stylised facts regarding our nationโ€™s largest asset class, which rose to six
trillion AUD in June 2016, making it three and a half times larger than the Australian share
market (Bleby & Greber 2016), but also recently has become increasingly topical, as
speculation builds that exponentially increasing housing demand, particularly in Sydney and
Melbourne is creating a bubble in the Australian housing market. In light of this it becomes
important to assess the role of the Australian housing market in monetary policy transmission,
Figure 1.1
Australian Housing Lending Rates
(Average interest rate on variable-rate loans)
Figure 1.2
Housing Prices and Interest Rates
(Reserve Bank of Australia 2016) (Reserve Bank of Australia 2014)
ย 
	
 ย 
	
 ย 
3
and also examine whether this has changed over the course of the last 25 years; former RBA
Governor Glenn Stevens flagged concern that the ability of monetary policy to influence
housing markets is today a channel that โ€œmay not be quite as effective as it once wasโ€ (Stevens
2015). Our work is further motivated by the absence in the existing literature of an Australian
analysis that considers the two possible components of the housing market transmission
mechanism separately, house prices and housing investment, rather than focusing solely on
house prices.
Our research is centred in analysing the effects of changes in the cash rate on house prices and
housing output; assessing the role of the Australian housing market as a channel of monetary
policy transmission. I conjecture the house price mechanism will behave as an immediate
effect, as housing stock remains fixed in the short term; while over the longer term the
investment channel will become more effective. Additionally, our analysis also addresses the
possibility of reverse feedback from both house prices and housing investment, to monetary
policy.
Examining the housing market as a transmission mechanism of monetary policy offers
potentially important policy implications. An understanding of this transmission mechanism in
an environment where its effectiveness is a concern of policy makers may allow the RBA to
more suitably conduct future monetary policy.
The results indicate the investment transmission mechanism is effective in transmitting cash
rate changes through the housing market. The house price channel does not operate as
intuitively expected; instead, house prices increase as a result of an increase in the cash rate.
We argue this is a particular instance of the generally observed โ€˜price puzzleโ€™ phenomenon.
The cash rate is found to operate exogenously from house prices and housing investment,
indicating the RBA does not set policy in direct response to the Australian housing market.
The remainder of this paper is structured as follows. Chapter two explores Australian monetary
policy and its transmission through the housing market. Chapter three provides a review of the
literature on monetary policy and housing markets. Chapter four describes the econometric
methods and data instrumental to our research. Chapter five presents our results. Lastly, chapter
six considers the policy implications of our findings.
ย 
	
 ย 
	
 ย 
4
Chapter Two โ€“ Australian Monetary Policy
and the Housing Market
To adequately assess the role of the Australian housing market, to the transmission of monetary
policy, a strong grasp of the operation of monetary policy, particularly in the Australian
inflation targeting context, and the nature of the Australian housing market is essential.
2.1 Monetary Policy, Then and Now:
The nature of current Australian monetary policy is a far cry from its original format as a central
banking function of the Commonwealth Bank. Since the Reserve Bank Act 1959 monetary
policy in Australia has been operated by a separate central bank authority, the RBA. A flavour
of the nature of monetary policy during this early period may be derived from Dr H.C โ€˜Nuggetโ€™
Coombs - Governor of the Commonwealth Bank (1949-1959) and RBA (1960-1968) -
endorsement of the view that,
โ€œA central bank should be like a good wife. It should manage its household competently
and quietly; it should stand ready to assist and advise; it can properly persuade and
cajole and on occasions even nag; but in the end it should recognise that the government
is the boss.โ€ (Coombs 1971 p63)
From this declaration comes the view that monetary policy was a tool to aid fiscal policy, rather
than operate within its own framework to achieve separate objectives. During the 1970s period
of stagflation, whereby stagnant economic conditions and high inflation were jointly present,
the Phillips curve trade off between unemployment and inflation disappeared, and the nature
of policy operation shifted greatly. Stephen Grenville declares it is from this period of
disruption that monetary policy was first appreciated as a โ€œseparate specialised function, with
ย 
	
 ย 
	
 ย 
5
a comparative advantage clearly distinguishable from fiscal policyโ€ (Grenville 2001),
necessitating a nominal anchor and framework to constrain the decision-making process of
monetary policy. Thus, the stage was set for the abandonment of money targeting, and the
adoption of inflation targeting, the framework under which monetary policy currently operates.
In the early 1990s the monetary policy objective of price stability was outlined by then RBA
Governor Bernie Fraser, as an average rate of inflation between two and three per cent over the
medium term, this explicit objective of the inflation targeting framework remains today. Figure
2.1 details the Australian inflation rate as measured by the proportional change in the Consumer
Price Index (CPI), extending from the late 1950s to the present day; illustrating that since the
introduction of the inflation targeting regime the proportional change in the CPI has in fact
remained on average within the two and three per cent band.
Figure 2.1
Australian Inflation
(Reserve Bank of Australia 2016)
2.2 Monetary Policy Transmission Channels:
The IS-LM model is the standard textbook model commonly used for monetary policy analysis,
in practise this model captures only one channel of monetary policy transmission; the interest
rate channel. In reality the transmission of monetary policy is far more complicated than the
operation of the interest rate transmission mechanism alone. The transmission channel
literature commonly identifies three further channels, the exchange rate channel, credit channel
and asset prices channel.
ย 
	
 ย 
	
 ย 
6
The four widely accepted transmission channels of monetary policy are discussed in detail by
Mishkin (1995). The interest rate channel, which is commonly demonstrated using the IS-LM
model, is regarded as the key mechanism in the basic Keynesian model, traditionally
recognised as primarily operating through business decisions about investment spending and
later recognised as operational through consumer decisions about housing and durable
expenditure. Taylor (1995) argues the importance of the interest rate channel in transmitting
effects of monetary policy to the economy, demonstrating that a contractionary policy shock
raises the short term nominal interest rate. The higher nominal interest rates therefore lead to a
decline in business fixed investment, residential housing investment, consumer expenditure on
durables and inventory investment, causing a decline in aggregate output and inflation.
The exchange rate transmission channel has, in the view of Mishkin (1995) become more
important to the understanding of monetary policy as globalisation increases. This channel
operates through the ability of the exchange rate to drive net exports. An increase in the
domestic interest rate increases the attractiveness of domestic deposits to foreign currency,
causing an appreciation of the domestic currency. By extension an appreciation of domestic
currency results in domestic goods being more expensive than foreign goods, thereby causing
a fall in net exports, aggregate output and inflation. Obstfeld and Rogoff (1995) emphasise the
importance of the exchange rate channel of monetary policy transmission and, in agreement
with Taylor (1995), noted that the framework for conducting monetary policy must be
international in scope, hence the consideration of the exchange rate transmission mechanism.
Bernanke and Gertler (1995) studied the credit channel of monetary policy transmission,
illustrating that two effects operate within the credit market to transmit monetary policy effects
to the economy, the bank lending effect and balance-sheet effect. The bank lending effect
operates upon a contractionary monetary policy shock, as the interest rate charged in the over
night money market increases, causing bank reserves to decrease, constraining lending and
thus reducing investment, and by extension aggregate output and inflation. The balance-sheet
transmission mechanism functions through the net worth of business firms. Contractionary
monetary policy causes prices to reduce, thus reducing the net worth of firms, resulting in
lenders having less collateral for their loans, reducing lending, and similar to the bank lending
effect, investment, aggregate output and inflation decreases
ย 
	
 ย 
	
 ย 
7
Lastly, the asset prices mechanism of monetary policy transmission. Meltzer (1995)
highlighted monetaristโ€™s objection to the Keynes framework for analysing monetary policy
effects on the economy as the focus is on only one asset price, the interest rate. Tobinโ€™s (1969)
q theory details the mechanism through which monetary policy affects the economy through
the value of assets. When monetary policy is contracted the public have restricted access to
money and reduce spending, decreasing the prices of assets such as equities on the stock
market. Lower equity prices under the q theory lead to a lower โ€˜qโ€™ value2
- the term
representative of the market value of firms divided by the replacement cost of capital - reducing
investment spending as it is expensive relative to equity value, which ultimately reduces
aggregate output and inflation. Alternatively, Modigliani (1971) argues the asset price
mechanism operates through wealth effects on consumption. Under this rationale when asset
prices fall the value of wealth decreases causing consumers to reduce consumption, and
ultimately as a result aggregate output and inflation declines.
Beyond the literature identifying these fundamental channels of transmission exists a vast body
of study regarding the effectiveness of monetary transmission to the real economy (Bernanke
et al 2005), cross country analysis (Musso 2011) and the relative importance of particular
traditional channels (Ibarra 2016).
2.3 The Australian Housing Market:
Having detailed the history of Australiaโ€™s inflation targeting monetary policy framework, and
the four common transmission channels, it may be acknowledged going forward, the
importance of specific asset prices, not only general inflation levels in maintaining economic
stability. While asset prices already serve a role in the inflation targeting framework, through
the CPI measure of inflation, experiences of many economies, most prominently perhaps, the
Unites States in 2008, and the enormous disruption asset bubbles cause (Miller & Stiglitz 2010)
has motivated many to consider whether monetary policy should do more to directly influence
asset prices. This conjecture leads us to a discussion of one of the most important asset markets
to the Australian economy, the housing market; an appreciation of which is fundamental to
2
If q is high the market value of firms is high relative to the replacement cost of its capital, new capital is cheap
relative to the market value of business. Companies may issue equity and receive a high price relative to the
capital purchased (Mishkin 1995).
ย 
	
 ย 
	
 ย 
8
understanding our assessment of its role in transmitting monetary policy throughout the
economy.
The Australian housing market is characterised by given housing stock in the short term and a
heightened sensitivity to interest rates, due to reliance upon borrowed funds. As a consequence
of lagged debt approval, land release, building permission and the generally long term nature
of housing production, supply response to monetary policy shocks is a delayed and long term
process. Due to these characteristics our analysis of the operation of the housing market in
response to monetary policy in the short term details a strikingly different story to a long term
analysis.
Figure 2.2 illustrates the response we expect of the housing market to an expansionary
monetary policy shock. On the vertical axis is house prices (PH), the horizontal axis measures
housing output (QH). A downward sloping curve (HD
) represents housing demand and a vertical
curve represents short term housing supply (HS
S.R). In the long run housing supply is gradually
more elastic and represented by an upward sloping curve (HS
L.R), as the housing supply curve
pivots clockwise over the time horizon. The model of the housing market features housing
prices and quantity as endogenous variables, the values of which are determined by the
intersection of housing demand and housing supply (short run or long run) curves. In the short
run housing quantity is determined by the vertical (inelastic) supply curve and is therefore
verifiably exogenous. Output becomes truly endogenous in the long run; a determinant of the
intersection of housing demand and long run housing supply curves, as supply is elastic and
responsive to house prices.
In the event of an expansionary policy shock demand for housing increases from HD
to HD1
, in
the short run as housing stock is fixed, prices in the market increase from HP0
t HP1
, while
output remains stable at HQ0
. In the longer term the housing supply curve gradually becomes
more elastic, as producers in the market are able to respond to the conditions of the market. In
the long term the same expansionary shock results in the price level only rising to HP*
, and
output increasing to HQ*
, as supply is capable of responding to increased demand due to lower
interest rates, thus reducing the pressure on the price level in the housing market.
ย 
	
 ย 
	
 ย 
9
Figure 2.2
Response of the Australian Housing Market to Expansionary
Monetary Policy
Unlike most goods markets, the housing market is not characterised in the short term by sticky
prices, and output is given in the short term. Thus, we conjecture that in the short run an
expansionary monetary policy shock will increase house prices, feeding through to increase
inflation. In the long run output is also affected, the direct price effect on inflation remains
positive, also operational is an indirect investment effect increasing inflation, aggregate
demand. Furthermore, we conjecture that the scale of these effects are relatively larger and that
the Australian housing market is more responsive to monetary policy shocks, than the average
goods market, due to the increased reliance upon borrowed funds, the majority of which are
variable rate mortgages.
2.4 Concluding Remarks:
Through discussion of the inflation targeting framework, a review of the transmission channels
and an analysis of monetary policy operation within the uniquely characterised Australian
housing market, we aim to have illustrated the nature of Australian monetary policy and the
housing market, through which policy effects inflation. With this general understanding, an
appreciation for the coming literature review regarding monetary policy and the housing
market may be formed. Furthermore, an elevated appreciation for the results of our
examination of the role of the Australian housing market, to the transmission of monetary
policy may be achieved.
ย 
	
 ย 
	
 ย 
10
Chapter Three โ€“ Literature Review
Recalling chapter one, our preoccupation is the effects of monetary policy upon house prices
and housing output, as we assess the role of the housing market in monetary policy
transmission, identifying the direct effect of house prices and the indirect effect of housing
investment (output) upon inflation. To this end we review literature identifying the effects of
monetary policy on output and inflation. A subset of this literature is empirical research, both
international and Australian, that analyses the response of the housing market to monetary
policy and we survey this specialisation. The principal monetary policy transmission channels
outlined in the literature have been surveyed previously in chapter two. Discussion regarding
the importance of asset prices to monetary policy has increased, particularly since the Global
Financial Crisis and recently, within Australia, due to the housing affordability problem in
certain capital cities. In light of this, and the possibility of reverse feedback from the housing
market to monetary policy, we review the asset price targeting literature. Lastly, as a precursor
to out results, we review literature identifying the price puzzle phenomenon. Common
empirical techniques used to assess the transmission of policy shocks to output and price levels
will be reviewed in the following chapter regarding modelling and data.
3.1 Monetary Policy Literature:
There exists a long empirical literature analysing how exogenous monetary policy shocks affect
output and inflation variables. This literature is classified under Monetarism, a school of
thought which asserts fairly definite predictions of monetary policy; that the money supply has
the ability to influence output and price levels, the foundation for empirical analysis from the
likes of Friedman and Fisher regarding monetary policy. Monetarists such as Friedman (1968)
theorise that irrespective of a central bankโ€™s targeting framework, monetary policy cannot
influence real variables such as unemployment and output growth over the long term, as shocks
only affect nominal variables, ie. the price level (inflation), income and interest rates, as per
ย 
	
 ย 
	
 ย 
11
the โ€˜liquidity effectโ€™ over an extended time horizon. Daniel (1981) is an example of early
empirical literature - extending upon Lucas (1972) - illustrating that announced contractionary
monetary policy in a small open economy holds a negative relationship with output, in the short
term. This result has become the general consensus view in empirical analysis, as the findings
of Sims (1980), Bernanke and Blinder (1992), Eichenbaum (1992) and Leeper and Gordon
(1992) all concur that an increase in the main tool of an inflation targeting central bank - the
policy interest rate - leads to a decline in output in the short term.
Shifting focus, we review literature formalising the effects of monetary policy upon inflation
as illustrated in our housing market model in chapter two. Irving Fisher in The Purchasing
Power of Money (1911) developed the relationship between changes in monetary policy and
changes in general price levels. Since this seminal paper, the literature has become vast. Haque
(1985) is representative of early literature, building upon the โ€˜Friedman-typeโ€™ beliefs held by
monetarists that monetary policy will not permanently influence real variables, although it
would exert an influence over inflation rates, especially in the long run. Haque illustrates with
contractionary monetary policy, steady state inflation rate decreases and inflation during the
transition path also falls. Furthermore, Karim et al (2011) and De Waal and Van Eyden (2014),
are examples of the modern literature examining the relationship between monetary policy and
inflation. Karim et al (2011) examines the price effects of monetary policy in the small open
economy of New Zealand; where contractionary monetary policy reduced inflation. De Waal
and Van Eyden (2014) conduct similar yet extended analysis on the South African economy,
investigating the impact of an increase in the official interest rate on both output and inflation,
noting Mishkinโ€™s (1995) reference to these elements as the โ€˜timing and effectโ€™ of monetary
policy on the economy. Interestingly, the response of inflation suggests a lag of approximately
24 months and evidence of the price puzzle โ€“ the positive relationship between tightened
monetary policy and inflation - observable in the first quarter, after which inflation declines.
Evident through this brief review of literature examining relationships between monetary
policy, output and inflation, is the consensus view that monetary contraction reduces inflation
and output. The consensus formalises our previous discussion of monetary policy operation;
the transmission channels reviewed in chapter two detail the means through which both effects
materialise.
ย 
	
 ย 
	
 ย 
12
3.2 International Housing Market Literature:
The means through which monetary policy affects output and inflation, are acknowledged as
transmission mechanisms of monetary policy; of which the most widely recognised have been
discussed at length in the previous chapter. The extension of this literature of greatest interest
to this dissertation, is the consideration of the importance of the housing market to the
transmission of monetary policy.
Internationally, research on the housing market as a transmission channel of monetary policy
has been carried out most densely in the United States (US) (Vargas-Silva 2008, Dore et al
2013, Christidou and Konstantinou 2011), while the United Kingdom (UK) (Elbourne 2005),
China (Wei et al 2014), South Africa (Gupta et al 2009) and developing economies such as
Turkey (Guler 2012) have also considered the housing marketโ€™s role in the transmission of
monetary policy to the economy, affecting price levels and output.
Empirical analysis of the importance of the housing market as a transmission mechanism of
monetary policy in the US has produced largely consistent results. Vargas-Silva (2008)
findings suggest that contractionary monetary policy shocks resulted in significantly negative
effects on housing starts and residential investment. The vector autoregressive (VAR) model
estimated by Christidou and Konstantinou (2011) over the period 1988-2009 similarly assessed
the effect of a monetary policy shock on US house prices and US housing investment. Again
contractionary monetary policy was found to lead to a reduction in housing investment.
Further, this effect in many states is found to be short lived and reversed within less than five
years; consistent with the theories of monetarists, that monetary policy is incapable of affecting
real variables in the long term. Further, a pronounced and long-lasting fall in housing prices is
documented. Analysis using an eight variable structural vector autoregressive (SVAR) model
of the UK executed by Elbourne (2005) again illustrated that a contractionary monetary policy
shock, resulted in decreased house prices.
Implementation of similar econometric analysis to South Africa and Turkey uncovered results
that further validate the US and UK analysis. Gupta et al (2009) finds South African house
prices increase in response to an expansionary policy shock, in consensus with the results of
previous literature. Empirical analysis of Turkey contributes to the consensus view that โ€œthe
ย 
	
 ย 
	
 ย 
13
interest rate affects the housing market considerably and house prices, particularly, play an
important role in the monetary transmissionโ€. (Guler 2012) The intuitive relationship between
monetary policy and house prices held before 2001; this relationship, however, is then
undermined, where results illustrate a comparable shock decreases house prices. The response
of housing investment is consistently compliant with expectations for the full period;
decreasing as monetary policy contracts.
Wei et al (2014) executed comparable analysis for the Chinese case and uncovered interesting
results that largely contradict the existing consensus; a contractionary monetary policy shock
appears to induce positive effects on both housing investment and house prices. The response
of prices is consistent with the price puzzle phenomenon; the literature upon which we will
expand upon later in our review.
3.3 Australian Housing Market Literature:
Domestically, fewer studies have examined the housing market and monetary policy, the
majority of which having been carried out over the last decade, as the GFC and the Australian
housing affordability problem, ignited the debate over whether monetary policy should respond
to the housing market, specifically house prices, see (Mishkin 2007). Within the literature, as
flagged by Wadud et al (2012), is a focus on identifying effects of monetary policy on house
prices and causes of overvaluation (see Fry et al, 2010), rather than an assessment of the
marketโ€™s importance as a transmission channel of monetary policy.
Wadud et al (2012) models the impact of monetary policy shocks upon the Australian housing
market. A nine variable SVAR model is estimated and impulse responses are generated for the
following variables; housing material costs, house prices, GDP, nominal domestic interest rate,
inflation rate, foreign interest rate, government spending on housing and the nominal exchange
rate against the dollar variables.3
The data, which includes quarterly housing approvals as a
proxy for housing investment and the Australian Bureau of Statistics (ABS) house price index
as a proxy for house prices, extends from 1974 - 2008.Wadud et al found that the short term
interest rate and inflation rate are the main determinants of house prices and house prices
significantly affect housing investment. Furthermore, the impulse responses, suggest a
3
The specification of the structural vector autoregression (SVAR) will be outlined in chapter four.
ย 
	
 ย 
	
 ย 
14
contractionary monetary policy shock has an immediate positive effect on housing investment
and significantly raise housing prices for a short period of time, both counter-intuitive to
expectations.
Phan (2014) uses four VAR models with differing strategies of identification to estimate the
contribution of private consumption and investment to the response of output to monetary
policy; each for Australia, US and UK over the period 1982-2007. The assessment of
Australian, US and UK investment and consumption in response to monetary policy, facilitates
a cross-country comparison of the composition of the output transmission mechanism of
monetary policy. Phan estimates that Australian housing investment accounts for 35 per cent
of the response of output, to monetary policy - far greater than the US or UK - and is cited as
a major cause of the differences observed between Australian, US and UK responses to
monetary policy shocks. The importance of the response of housing investment to the
difference of the transmission of monetary policy is a conclusion that validates our assessment
of housing investment as a separate channel of transmission within the housing market; in
addition to the house price channel which is far more present in the existing literature.
3.4 Asset Price Targeting Literature:
Literature detailing the effects of monetary policy upon housing output and prices has been
discussed at length. However, this relationship has been observed in the reverse, whereby house
prices inform monetary policy. This body of literature raises two questions, the first, whether
central banks, should respond to asset price signals. The second, and most relevant to our
research, is not a matter of whether central banks should respond โ€“ a normative question โ€“ but
instead a positive question of whether monetary policy does respond to asset prices. The former
is dealt with largely through theoretical literature, while whether central banks operate
reactively to asset price signals is found in empirical literature.
Literature discussing the validity of asset prices as informants of monetary policy and the view
that monetary policy should react to asset price misalignments has developed rapidly in recent
years. However, the idea that central banks need to account for asset prices in monetary policy
is not a new phenomenon, first proposed by Irving Fisher who believed policy makers should
try to stabilise a broad index of asset prices such as property, as well as the traditional consumer
ย 
	
 ย 
	
 ย 
15
goods and services index. Poole (1970) addressed potential benefits, discussing the โ€˜leaning
against the windโ€™ monetary policy strategy that advocates changing interest rates when
disturbances originate in asset markets. Kent and Lowe (1997) constructed a model that
incorporated the notion of asset price misalignments into monetary policy, while Cecchetti et
al (2002) argued that theoretical reasons exist for believing an inflation targeting central bank
โ€“ such as the RBA โ€“ may improve macroeconomic performance by reacting to asset price
misalignments. Zhao and Gao (2010), as a consequence of their Chinese analysis,
recommended that asset prices be considered as endogenous variables in the monetary policy
function, in order to make the objectives of monetary policy more controllable by the central
bank. This conclusion is highly controversial and disagreed with by Bernanke and Fertler
(2001), in part due to the difficulty in interpreting asset price misalignments.
In addition to the debate over the merits of asset price targeting, papers have examined whether
central banks already respond to asset prices. Zhao and Gao (2010) examined the impact of
asset price fluctuation on Chinaโ€™s monetary policy, using quarterly data during the period of
1994-2006. The results from Granger causality and cointegration tests showed for every one
per cent increase in housing prices, monetary policy interest rates increased by approximately
two percent. Fiodendji (2012) examined whether Canadian monetary policy operates in
response to asset prices. While the Canadian central bank admits consideration of asset prices
in policy deliberations, due to their contribution to inflation, there has been explicit denial of
an effort to directly stabilise asset prices. However, the findings suggest that stock market
stabilisation plays a larger role than has been acknowledged. Additionally, Castro and Sousa
(2010) found evidence of housing price accommodation in the monetary policy of both the US
and UK. Whether reverse feedback exist between the Australian housing market and monetary
policy, or rather whether the housing market โ€“ specifically house prices โ€“ influences monetary
policy, will contribute to this growing literature.
3.5 Price Puzzle Literature:
Forthcoming results from our research indicate the existence of a sectoral example of the price
puzzle phenomenon, operating within the Australian housing market. We now move to briefly
review the existing price puzzle literature, developing an understanding of how our research
compares and contributes to the existing discussion. The term โ€˜price puzzleโ€™ was first coined
ย 
	
 ย 
	
 ย 
16
by Eichenbaum (1992) in his commentary on Simsโ€™ (1992) seminal paper studying the effects
of monetary policy across countries, which identified in a SVAR model that a contractionary
monetary policy shock โ€“ an increase in the federal funds rate โ€“ was associated with a counter-
intuitive, persistent increase in the price level, ie the operation of a โ€˜price puzzleโ€™. Since Simsโ€™
discovery and Eichenbaumโ€™s formalisation of the price puzzle, a vast body of literature has
developed. It has been reported that approximately 50 per cent of modern studies using VAR
analysis to investigate the effects of monetary policy report that after a policy contraction prices
increase, at least in the short run. (Rusnak et al 2013) One of the main explanations of the price
puzzle is model mis-specification, particularly due to omitted variables, see (Rusnak et al
2013).
Evidence of the price puzzle has been uncovered in both developed eg. US (Hanson 2004) and
developing economies eg. Pakistan (Javid & Munir 2010), across the aggregated economy and
within specific markets eg. the Chinese housing market (Yixun 2010) and the Australian
housing market, as flagged in the findings of Wadud et al (2012) and Phan (2014). Further,
research has been conducted into resolving the puzzle, focusing on how best to specify models
to remove or reduce the puzzle. However, no formal, widely applicable approach has been
developed and, no universal explanation for the puzzle exists. In sum, the price puzzle, remains
as puzzling as ever. Our research contributes to the evidence of this anomalies existence.
3.6 Contribution to the Existing Literature:
Our research fills a gap in the Australian literature where assessment of the role of the housing
market in transmitting effects of monetary policy to inflation is sparse. While exhibiting
similarity to Wadud et al (2012), in implementing a SVAR model to assess the response of the
housing market to monetary policy, our analysis is differentiated by the intention to capture the
effects of monetary policy on house prices and housing output, implementing a more focused
SVAR that adopts fewer variables. Previous studies - particularly in the Australian context -
have not directly assessed the importance the of housing market as a separate transmission
mechanism of monetary policy, rather than a contributor to the interest rate and credit channels
of transmission. Further differentiation comes from our focus in deciphering the role of each
effect of monetary policy, on both house prices and housing investment, in maintaining
ย 
	
 ย 
	
 ย 
17
inflation. Lastly our consideration of possible reverse causation will serve to contribute to the
growing debate over asset prices as monetary policy targets.
ย 
	
 ย 
	
 ย 
18
Chapter Four โ€“ Method and Data
Following our review of the literature this chapter will outline the econometric method we
follow in assessing the effects of monetary policy on house prices and housing output. This
discussion involves explanation of ordinary least squares (OLS), the structural vector
autoregressive (SVAR) model and the Cholesky identification process, among other
econometric particulars. Furthermore, the data employed, in representation of several variables
within the SVAR model, is outlined.
4.1 Ordinary Least Squares Method (OLS):
OLS estimation of the relationship between monetary policy and the Australian housing
market, facilitates preliminary analysis and provides an initial sense of the results. The
relationship of house prices and output with monetary policy is estimated through the following
pair of equations:
OLS operates under the assumption of exogeneity of the independent variable ie. cash rate, and
ignores endogeneity between variables. Endogeneity is fundamentally important to the
reliability and accuracy of the relationships estimated, as well as the validity of statistical tests
and also is relevant to our secondary concern of the housing market informing monetary policy.
This necessitates the progression to a SVAR model, in which all variables are endogenous, so
it is not reliant upon the assumption of exogeneity.
๐ป๐‘œ๐‘ข๐‘ ๐‘’	
 ย  ๐‘ƒ ๐‘Ÿ๐‘–๐‘๐‘’๐‘ 	
 ย  = 	
 ย  ๐›ผ + ๐›ฝ๐ถ๐‘Ž๐‘ โ„Ž	
 ย  ๐‘… ๐‘Ž๐‘ก๐‘’	
 ย  + ๐‘’
๐ป๐‘œ๐‘ข๐‘ ๐‘–๐‘›๐‘”	
 ย  ๐‘‚ ๐‘ข๐‘ก๐‘๐‘ข๐‘ก = 	
 ย  ๐›ผ + 	
 ย  ๐›ฝ ๐ถ๐‘Ž๐‘ โ„Ž	
 ย  ๐‘… ๐‘Ž๐‘ก๐‘’ + ๐‘’
(4.1)
(4.2)
ย 
	
 ย 
	
 ย 
19
4.2 Structural Vector Autoregressive Model (SVAR):
SVAR models are adopted in the overwhelming majority of research assessing monetary policy
transmission. Sims (1980) introduced the use of VAR models for policy assessment. Prior to
this formative paper the tool was used in forecasting, while simultaneous equation models were
the mainstream method of economic analysis. The VAR model is used in economic analysis
due to possible endogeneity of variables such as monetary policy โ€“ the cash rate โ€“ that simple
OLS cannot account for. Thus, modelling is necessary for accurate and reliable analysis. Within
the literature different forms of VAR are adopted, reduced form VAR, structural (SVAR) and
cointegrated VAR models being the most common. The dominant use of the VAR model
within the literature, to assess monetary policy transmission is for the most part uncontested,
with the exception of the dynamic stochastic general equilibrium (DSGE) model.
The DSGE model has been employed sparingly throughout the literature by the likes of
Christidou & Konstantinou (2011) and Funke et al (2015). Although DSGE has been
previously effective, the method is not ideally suited to empirical policy evaluation. In light of
this the VAR model will be employed in our assessment of the importance of the Australian
housing market to the transmission of monetary policy. The VAR model allows for dynamic
feedback between variables and is useful in identifying effects from exogenous shocks such as
monetary policy shocks. It is clear from the summary of existing literature which implemented
VAR models, that the particular VAR form used is dependant on the nature of the data
representing the variables of the model. In the case of our research, the specific VAR models
considered are the SVAR and vector error correction (VEC) models. Whether the data sets
exhibit evidence of non-stationarity and cointegration, determines the appropriateness of either
the SVAR or VEC models.
As the subsequent stationarity and autocorrelation results conclude, the estimation of a SVAR
model is appropriate for our analysis. Our model takes the form of a three-variable SVAR
system of the cash rate (CR), house prices (HP) and housing output (HQ) and may be expressed
as per equation (4.3):
	
 ย 
ย 
	
 ย 
	
 ย 
20
๐ถ๐‘…8 =	
 ย  ๐›ฝ9 + ๐›ฝ:;
<
;=9
๐ถ๐‘…8>; +	
 ย  ๐›ฝ?;
<
;=9
๐ป๐‘ƒ8>; +	
 ย  ๐›ฝ@;
<
;=9
๐ป๐‘„8>; +	
 ย  ๐œ–C8	
 ย 
๐ป๐‘ƒ8 =	
 ย  ๐›พ9 + ๐›พ:;
<
;=9
๐ถ๐‘…8>; +	
 ย  ๐›พ?;
<
;=9
๐ป๐‘ƒ8>; +	
 ย  ๐›พ@;
<
;=9
๐ป๐‘„8>; +	
 ย  ๐œ–E8	
 ย 
๐ป๐‘„8 =	
 ย  ๐›ฟ9 + ๐›ฟ:;
<
;=9
๐ถ๐‘…8>; +	
 ย  ๐›ฟ?;
<
;=9
๐ป๐‘ƒ8>; +	
 ย  ๐›ฟ@;
<
;=9
๐ป๐‘„8>; +	
 ย  ๐œ–G8
(4.3)
4.2.1 Identification of Monetary Policy Shock:
While the merits of the VAR model are generally undisputed in the literature, the method for
identification of the monetary policy shock is decidedly less unanimous. Ramey (2011) argued
that in standard VAR shock-identification procedures, shocks are partially predictable and
cannot be considered to be unexpected as assumed in theoretical analysis. Following Simsโ€™
(1980) seminal paper, and in line with Rameyโ€™s concern, many papers employ VAR models
with different identification strategies.
A common approach used to solve this problem, is the Cholesky decomposition, first proposed
by Sims (1980). The Cholesky decomposition orthogonalises the shocks, including shocks to
monetary policy, to recover the structural errors. However, with the Cholesky decomposition
ordering of the variables is critically important; we account for this by repeating our model
with alternative variable ordering in the sensitivity analysis of our results.
4.2.2 Impulse Response Functions:
From our SVAR model we generate impulse response functions (IRF)s. An IRF graphically
traces the temporal responses of each endogenous variable to an exogenous positive shock in
an error, such as the one designed to capture monetary policy. We impose this shock as a one
standard deviation increase in the monetary policy variable and include 2 standard errors and
95 per cent bootstrapped confidence intervals. IRFs allow graphical analysis of the direction,
magnitude and persistence of the temporal responses to a monetary policy shock, thus enabling
close examination of the relationships between monetary policy and the Australian housing
market.
ย 
	
 ย 
	
 ย 
21
4.2.3 Block Exogeneity Wald Test:
Upon estimation of our SVAR model and generation of IRFs, all variables are treated as
endogenous. To formally assess the possibility of monetary policy acting in response to the
Australian housing market we test the block exogeneity of the model, with respect to the cash
rate; conducting a Wald test of block exogeneity which treats a previously endogenous variable
- the cash rate - as exogenous and tests for joint significance of each of the other endogenous
variables in the cash rate equation. The null hypothesis is the exogenous variable is only acting
in response to lagged versions of itself. We specify this test algebraically in the forthcoming
chapter, accompanied by discussion of its findings. Through this test we determine whether the
cash rate, is in fact responding to the housing market or acts independently of these variables.
4.3 Variables:
Having outlined the method this dissertation will follow, we now outline the variables
implemented within the SVAR model. Existing literature has implemented an array of
variables. However, the housing market model we derived in chapter two, relied only on the
variables of house prices and housing output, in addition to the cash rate variable. Therefore,
following the model of the housing market and in an effort to narrow the focus of our model
we start from the minimum specifications necessary. As we cannot say anything sensible about
the importance of the housing market to the transmission of monetary policy without including
the three variables of most central relevance to the housing market model, the cash rate, house
prices and housing output will be implemented, thus creating a three variable SVAR.
We acknowledge different models will suggest the implementation of different additional
variables; however, no one model may satisfy them all simultaneously, hence we reserve
additional variables for implementation in sensitivity testing of the results of our three-variable
SVAR model. These additional variables will include the Australian โ€“ Chinese exchange rate,
growth of disposable income, commodity price index and a building cost variable.
ย 
	
 ย 
	
 ย 
22
4.4 Data:
Our research is concerned with the Australian case of housing market monetary policy
transmission and will therefore utilise Australian data. The data employed in our model spans
the period 1990Q1 - 2015Q4, governed by the widespread consensus that operation of
Australiaโ€™s central bank inflation targeting regime is observable from the early 1990s onward
(Fraser 1993), while low inflation targets have been a feature of RBA rhetoric since as early as
1989 (Stevens 1999). For monetary policy, the RBA monthly cash rate target (CR) data is used,
this data is available from 1990Q3 onward, again dictating our period of analysis. We
extrapolate this data using the 90-day bank accepted bill rate for the two prior quarters to
provide a rounded 25-year period. The use of CR is comparable to existing literature and is a
particularly good fit for our research as the RBA uses the cash rate as its sole instrument, thus
movement of the cash rate is fully illustrative of monetary policy.
The Australian Bureau of Statistics (ABS) Consumer Price Index (CPI) is used to extract data
representative of Australian house prices; as the purveyor of official housing market
information, it is regarded as authoritative (Joye 2016). The data set includes an index of
housing prices weighted across the eight capital cities: Adelaide, Brisbane, Canberra,
Melbourne, Perth and Sydney and is recorded on a quarterly basis. Further, this data was also
implemented in Wadud et al (2012) as representative of Australian house prices. The house
price index (HPI) weights fluctuate throughout the 25-year period, yet new housing
consistently accounts for over 70 percent; rental costs complete the index. The ABS
Residential Property Price Index is not used as it only extends back to 2002.
Housing output in our preferred model is represented by the ABS dwelling commencement
(DC) data. This data records the number of new dwellings commenced each quarter within the
private sector; both houses and other types of residential property ie. residential apartments.
In our sensitivity analysis we make use of additional data sets. Firstly, an implicit price deflator
(IPD) of housing prices is sourced from the ABS and used as an alternative to the CPI house
price index. Further, gross fixed capital formation chain weighted index, with respect to new
housing (GFCF) is used as an alternative to dwelling commencements, as is dwelling approvals
(DA) which records the number of residential dwellings, both houses and other residential
dwellings approved for construction. Data on housing finance commitments (FC) from the
ย 
	
 ย 
	
 ย 
23
ABS, capturing the value of finance both owner occupiers and investors accrue in purchasing
new dwellings, is used as an additional proxy for housing output. Further, we examine the pass
through of CR to interest rates, using the mortgage rate (MR) as alternative to the cash rate.
MR data comes from the RBA; we use the standard variable owner occupier rate and believe
it is appropriate due to Australiaโ€™s unique composition of mortgages, an overwhelming
majority of which are variable rate mortgages.
In addition to alternative data sets, we further our sensitivity analysis to reduce omitted variable
bias, including a number of additional variables. We include a variable representative of the
cost of building new homes (BC), using the ABS value of work data set to account for the
increasing wages and housing materials. This is in line with the ABS approach to accounting
for increasing costs of home building. Disposable income growth (YdG) data sourced from the
ABS is included to account for increasing Australian disposable income. Further, we include
data tracing the Australian to Chinese exchange rate (ACX), to examine the effect of Chinese
real estate investment to house prices and housing outputs response to monetary policy. Lastly,
in alignment with existing literature, we include a domestic commodity price index (CI) data
from the RBA, in an effort to capture the forward looking rationale of monetary policy. All
data is seasonally adjusted where available. Where appropriate, we convert data into its natural
logarithm, transforming skewed distributions into more symmetrical distributions, and
reducing scale discrepancies. We do not take the natural logarithm of CR or MR as they are
rates.
4.4.1 Stationarity:
The use of non-stationary data in SVAR models may produce unreliable results, as the standard
error bounds of SVAR IRFs assume a stationary and normal distribution, while non-stationary
data do not have the same distribution; ultimately leading to poor or incorrect understanding.
Therefore, it is important to establish whether our data is stationary or non-stationary, and
correct for non-stationary data. A preliminary manner of investigating stationarity is plotting
each of the variables over time to observe whether the mean and variance appear dependent on
time. In Figure 4.1 we graph CR, HPI and DC as a preliminary measure of helping fit our data.
ย 
	
 ย 
	
 ย 
24
Figure 4.1
Cash Rate, ln(House Price Index), ln(Dwelling Commencements) Plots
Observable within the CR plot is the long period of monetary expansion that was enacted in
the 1990s, subsequent to the announcement of the inflation target of two - three percent, and in
response to the economic downturn of the early 1990s. This decline is observable until 1994.
After six years of contraction, CR reduced in response to the GFC to what was then their lowest
ย 
	
 ย 
	
 ย 
25
level in history, three per cent. Subsequently, rates climbed again, peaking in 2011 at 4.75 per
cent, then declining in an attempt to stimulate the economy which is yet to return to its pre
GFC pace of activity, sitting at the close of 2015 at one and a half per cent.
In examining the HPI plot we observe cyclical increases and decreases within the period 1990-
2000. However, this cyclical pattern is broken when HPI begins its strong, consistently upward
trajectory. This unprecedented increase was not restricted to Australia only, but was rather, a
worldwide phenomenon. The RBA has labelled the 2000s as when โ€œhousing put itself on the
national agendaโ€ (Yates 2011); from this graph we can clearly see why the debate surrounding
house prices and the house price bubble question remains as fiery as ever. The housing bubble
problem was resolved for many economies after the GFC, as house prices dropped; however,
this was not the case in Australia as HPI continues to steadily increase.
DC shows a greater degree of volatility than HPI, yet the upward trend is similarly evident as
DC displays an upward trend since 2012. The decreasing CR in the early 1990s are evident, as
DC increased consistently. The effect of the introduction of the GST on the 1st
July 2000 is
also apparent, as housing investment was brought forward, causing a significant reduction in
DC in the second half of 2000 (Kearns & Lowe 2011). The effects of the GFC are also apparent,
accounting for the significant drop in 2008 that was subsequently reversed as a consequence
of the contractionary monetary policy.
In addition to revealing economic context, these plots are informative starting points in
assessing stationarity. Upon initial examination HPI is expected to be non-stationary while DC
may be stationary, CR could be stationary or perhaps a random walk with a negative drift.
4.5 Primary Variable Tests:
To validate these ad hoc preliminary assessments, we move to a formal econometric
stationarity assessment for which the augmented Dickey Fuller test (ADF) is used. The ADF
test, pioneered by Dickey and Fuller (1981), is represented algebraically by equation (4.4):
ย 
	
 ย 
	
 ย 
26
โ–ณ ๐‘ฆ8 =	
 ย  ๐›ผJ + ๐›ฝ๐‘ก + 	
 ย  ๐›พ ๐‘ฆ8>9 +	
 ย  ๐›ฟฮ”๐‘Œ8>;M9
N
;=:
+	
 ย  ๐œ€8
The ADF test tests the null hypothesis H0: ฮณ = 0, meaning the time series has a present unit
root, and is thus non-stationary. If the null hypothesis is rejected in favour of the alternative
hypothesis HA: ฮณ โ‰  0, it is concluded that there is no unit root present and the time series is
stationary. The ADF test is calculated across two deterministic specifications, intercept alone,
assessing whether the series is a random walk around a drift or stationary, and an intercept and
trend assessing whether the series is a random walk, or is stationary around an underlying
deterministic trend. We calculate the ADF test at a four lag maximum, given the quarterly
nature of our data. The p-values from the results are reported in Table 4.1; we accept or reject
the null hypothesis at the five per cent significance level.
Table 4.1
Unit Root Tests
Level Data First Differences
Lag Intercept Intercept + Trend No Intercept Intercept
CRt 1 0.0163 0.0107
2 0.0086 0.0039
3 0.0032 0.0016
4 0.0165 0.0035
ln(HPIt) 1 0.9978 0.2621 0.0013 0.0049
2 0.9858 0.1421 0.0018 0.0059
3 0.9898 0.1070 0.0170 0.0489
4 0.9684 0.1446 0.0046 0.0087
ln(DCt) 1 0.0673 0.0475
2 0.0458 0.0374
3 0.0225 0.0127
4 0.2710 0.2631
The results of the ADF test on CR demonstrate that the null hypothesis should be rejected at
all lags, across both deterministic specifications of the test; therefore, no unit root exists. CR is
neither a random walk around a constant drift nor a random walk around a deterministic trend;
the series is stationary.
We fail to reject the null hypothesis when testing HPI, across all lags and both deterministic
specifications, thus we conclude that in levels HPI is a non-stationary series. Moving to test
HPI in first difference form, we can reject the null hypothesis, although not consistently across
all lags and specifications. In first differences with an intercept, the null hypothesis is
(4.4)
ย 
	
 ย 
	
 ย 
27
consistently rejected, thus with a constant drift, no unit root is present and HPI is stationary in
first differences.
When testing DC in levels under both deterministic specifications, we reject the null hypothesis
inconsistently across the four lags. As the ADF test is predicated on the absence of
autocorrelation, we move to test for the presence of autocorrelation in the residuals of the ADF
equation at various lags. The Breusch-Godfrey serial correlation LM test is used to conduct
this analysis. We test for autocorrelation of the first-fourth order as a consequence of our
quarterly data. Under the same reasoning we continue to test to four lags. As reported by the
p-values in Table 4.2, we may reject the null hypothesis of zero autocorrelation for lags 1 and
3 when an intercept is imposed and at lags 1 and 2 when a trend is also included; failing to
reject the null hypothesis at the remaining lag specifications. In the ADF test at lag 2, with a
constant and lag 3, with both a drift and deterministic trend we reject the null hypothesis of the
presence of a unit root. Therefore, where we accept the Breusch-Godfrey null hypothesis of
zero autocorrelation, and also reject the ADF null hypothesis of a present unit root, we conclude
that DC is stationary.4
Table 4.2
Breusch-Godfrey Serial Correlation LM Test
We have concluded that two variables, (CR and DC) are I(0), stationary and one (HPI) is I(1),
non-stationary and therefore, cointegration is impossible. We correct for the I(1) variable by
using the first differences of the series, estimating our primary SVAR model with CR, lnDC
and the first difference of lnHPI. Estimating a SVAR model relies on the assumption of
stationary time series data, we can now confirm that our primary model fulfills this assumption.
4
Since this conclusion is a relatively close call we will re-specify the model using the first differences of
dwelling commencements, analysing the accumulated responses, comparing if the results are similar when using
dwelling commencements in levels.
Level Data
Lag Intercept Intercept + Trend
ln(DCt) 1 0.0074 0.0112
2 0.1044 0.0342
3 0.0423 0.0520
4 0.3415 0.2697
ย 
	
 ย 
	
 ย 
28
4.6 Robustness Variables Preliminary Statistics:
4.6.1 Alternative Data Sets
In addition to the data of our preferred model, we employ several alternative data sets in our
robustness testing. These are the mortgage rate (MR), implicit price deflator (IPD), dwelling
approvals (DA), gross fixed capital formation index (GFCF) and finance commitments (FC).
We now test the stationarity of these data using the same method followed previously. The
ADF test is used to test the presence of a unit root in series, the p-values are reported in Table
4.3 and the Breusch-Godfrey test considers the existence of autocorrelation in the residuals,
the p-values from which are reported in Table 4.4.
Table 4.3
Unit Root Tests
Level Data First Differences
Lag Intercept Intercept + Trend No Intercept Intercept
MRt 1 0.0057 0.0113
2 0.0004 0.0018
3 0.0100 0.0363
4 0.0133 0.0455
ln(IPDt) 1 0.9943 0.5624 0.0006 0.0000
2 0.9908 0.4688 0.0084 0.0004
3 0.9888 0.3924 0.0391 0.0062
4 0.9839 0.3509 0.0849 0.0212
ln(DAt) 1 0.0096 0.0083 0.0000 0.0000
2 0.1895 0.2279 0.0000 0.0000
3 0.1876 0.2299 0.0000 0.0002
4 0.0059 0.0066 0.0000 0.0000
ln(GFCFt) 1 0.4997 0.0026 0.0000 0.0000
2 0.6042 0.1091 0.0000 0.0001
3 0.4876 0.0476 0.0000 0.0000
4 0.6399 0.1514 0.0000 0.0001
ln(FCt) 1 0.3210 0.3113 0.0000 0.0000
2 0.3225 0.4739 0.0000 0.0000
3 0.1731 0.4849 0.0000 0.0000
4 0.1167 0.5189 0.0008 0.0002
Similarly, to CR, we reject the ADF null hypothesis for all lags and across deterministic
specifications of MR, thus MR is stationary in its level form.
ย 
	
 ย 
	
 ย 
29
When testing IPD the null hypothesis of the ADF test cannot be rejected across all lag levels,
regardless of whether a drift or trend is imposed. When we first difference the series we reject
the null hypothesis inconsistently when no constant drift is imposed, yet when we impose a
constant drift the null hypothesis of a present unit root can be consistently rejected. Therefore,
we conclude when in first differences and a constant drift is imposed, the IPD series is
stationary.
We reject the ADF null hypothesis inconsistently across lags and deterministic specifications
for DA, when in levels. We continue to test the DA series in levels under the Breusch-Godfrey
test, testing the hypothesis of no first-fourth order residual autocorrelation. We reject for all lag
levels except for the second, when a constant drift is imposed on the series. This lag and
deterministic specification corresponds to a non-rejection of the ADF null hypothesis, therefore
we conclude DA is non-stationary in levels. Moving to first differences, we can conclusively
reject the ADF null hypothesis at all lags levels and deterministic specifications, therefore we
conclude DA is stationary in first differences.
The presence of a unit root in GFCF, cannot be rejected consistently across lags and
deterministic specifications when the series is in levels. Examining autocorrelation in the
residuals, we fail to reject the null hypothesis for all lags when a drift is imposed, except for
the second, and cannot reject at any lag when a trend is also imposed. These lags correspond
to ADF results that reject the null hypothesis of the ADF test, therefore we conclude GFCF is
stationary in levels.5
When examining FC in levels we consistently fail to reject the ADF null hypothesis, therefore,
in levels across lags and deterministic specifications the series is non-stationary. However, in
first differences we consistently reject the ADF null hypothesis and conclude that FC is
stationary, when in first differences regardless of whether a constant drift is imposed or not.
5
Again, we will look at the first differences in accumulates responses to confirm the validity of our conclusion
since it is somewhat of a close call.
ย 
	
 ย 
	
 ย 
30
Table 4.4
Breusch-Godfrey Serial Correlation LM Test
Level Data First Differences
Lag Intercept Intercept + Trend No Intercept Intercept
ln(IPDt) 1 0.4791 0.3656 0.0579 0.8395
2 0.8689 0.6531 0.2242 0.9192
3 0.8955 0.7938 0.6737 0.9661
4 0.9421 0.9650 0.8647 0.9471
ln(DAt) 1 0.0004 0.0004 0.0048 0.0056
2 0.0970 0.0433 0.0120 0.0136
3 0.0023 0.0023 0.0038 0.0038
4 0.0192 0.0145 0.0036 0.0038
ln(GFCFt) 1 0.0647 0.1483 0.3504 0.2278
2 0.0002 0.3160 0.4622 0.2766
3 0.2214 0.3823 0.9058 0.7210
4 0.4059 0.3984 0.6473 0.4112
4.6.2 Additional Variables
In addition to the data in our preferred model, we employ several additional variables in our
robustness testing of our primary results. These are disposable income growth (YdG), building
costs (BC), Australian-Chinese exchange rate (ACX) and commodity price index (CI). We now
test the stationarity of these data, the p-values from the ADF and Breusch-Godfrey tests are
presented in Table 4.5 and 4.6, respectively.
The ADF testing of YdG in level form, concludes that the null hypothesis is rejected across all
lags and deterministic specifications. Therefore, YdG is stationary in levels.
The ADF null hypothesis cannot be rejected consistently across deterministic specifications for
BC when in levels. Looking to the Breusch-Godfrey test, we reject the null hypothesis of zero
first-fourth order autocorrelation for all lags. When we transform BC into first differences, we
consistently reject the ADF null hypothesis of non-stationarity across all lags and both
specifications; concluding that BC in first differences is a stationary series.
Both the ACX and CI series, when in levels consistently cannot reject the ADF null hypothesis,
yet in first differences the null hypothesis is consistently rejected across all lags and both
deterministic specifications. Therefore, in first differences, the ACX and CI series are
stationary.
ย 
	
 ย 
	
 ย 
31
Table 4.5
Unit Root Tests
Level Data First Differences
Lag Intercept Intercept + Trend No Intercept Intercept
ln(YdGt) 1 0.0000 0.0000 0.0000 0.0001
2 0.0000 0.0000 0.0000 0.0000
3 0.0001 0.0003 0.0000 0.0000
4 0.0070 0.0397 0.0000 0.0000
ln(BCt) 1 0.1852 0.0113 0.0000 0.0000
2 0.3123 0.0436 0.0000 0.0000
3 0.5165 0.2461 0.0000 0.0003
4 0.1171 0.0046 0.0000 0.0000
ln(ACXt) 1 0.0567 0.2979 0.0000 0.0000
2 0.0802 0.3639 0.0000 0.0000
3 0.0826 0.3660 0.0000 0.0000
4 0.1115 0.4477 0.0000 0.0000
ln(CIt) 1 0.5573 0.2585 0.0000 0.0000
2 0.6476 0.6054 0.0000 0.0000
3 0.7070 0.6214 0.0000 0.0000
4 0.7200 0.8788 0.0000 0.0000
Table 4.6
Breusch-Godfrey Serial Correlation LM Test
Level Data First Differences
Lag Intercept Intercept + Trend No Intercept Intercept
ln(BCt) 1 0.0002 0.0000 0.0004 0.0005
2 0.0004 0.0000 0.0000 0.0000
3 0.0006 0.0003 0.0001 0.0001
4 0.0003 0.0007 0.0002 0.0001
In this chapter we looked at the base SVAR model, which we estimate in terms of CR, HPI and
DC. We found HPI to be non-stationary, while DC and CR were both stationary. Therefore,
we specify our base SVAR model in terms of CR, lnDC and the first difference of lnHPI. In
addition, we specified the additional data and variables that will be used for sensitivity analysis.
We found GFCF, MR and YdG to be stationary, whereas DA, IPD, FC, BC, AXC and CI are
non-stationary and are specified in first differences to correct for this. Having determined our
data is stationary, either in levels or first differences; we conclude the primary SVAR model
and sensitivity analysis is fit for purpose. We examine our results in the forthcoming chapter.
ย 
	
 ย 
	
 ย 
32
Chapter Five - Results
Having reviewed the existing literature, and the common empirical structural vector
autoregression (SVAR) technique, we now combine the SVAR method and data described in
the preceding chapter to analyse the responses of the Australian housing market to monetary
policy. Our primary preoccupation is to examine the effects of monetary policy on house prices
and housing output; effectively assessing the role of the Australian housing market in monetary
policy transmission. Secondarily we seek to determine whether evidence exists to suggest
monetary policy operates in response to the housing market.
5.1 Preliminary Analysis:
5.1.1 Ordinary Least Square Regression Analysis
Elementary ordinary least squares (OLS) regression analysis provides us with an indication of
the relationship between the cash rate and the housing market we may anticipate in our more
sophisticated SVAR analysis.
In Table 5.1 Panel A we report the results of two OLS regressions, equations (4.1) and (4.2);
having individually regressed the house price index (HPI) and dwelling commencements (DC)
on the cash rate (CR), from zero to two lags. HPI shows a positive relationship with CR from
zero to one lag; at the second lag the relationship becomes negative and is our first significant
result.6
Further, the magnitude of the negative relationship in the second lag is approximately
equivalent to the cumulative magnitude of the positive relationships between the
contemporaneous cash rate and the cash rate lagged one period. This suggests that HPI at first
responds positively to CR; however, by the second lag this relationship becomes negative and
6
We continue to use the five per cent level of significance.
ย 
	
 ย 
	
 ย 
33
approximately outweighs the magnitude of the positive relations. This result is unexpected,
suggesting HPI responds counter-intuitively at first to CR, yet when lagged by two periods this
is corrected for, by a large and intuitive (negative) response.
Table 5.1
OLS Regression Results
Dependent Variable
Independent Variable House Price Index (HPI) Dwelling Commencements (DC)
Panel A Lag Coefficient P- Value Coefficient P -Value
Cash Rate (CR) - 0.006638 (0.0780) 1.143227 (0.2557)
Cash Rate (CR) 1 0.001578 (0.8063) -3.012966 (0.0033)
Cash Rate (CR) 2 -0.008213 (0.0164) 3.968290 (0.0001)
Panel B
Cash Rate (CR) 1 0.011893 (0.0000) -0.078672 (0.0000)
Cash Rate (CR) 2 -0.012221 (0.0000) 0.066659 (0.0000)
Examining the results of this regression with respect to DC, it appears, that housing output
responds intuitively (negatively) to CR quicker than house prices, as the result at lag 1, is not
only significant, but negative and large in magnitude. Interestingly, at lag 2, this response
appears to be outweighed by a larger positive response that is also significant.
The unexpected relationship between HPI and CR could perhaps be the result of potential
endogeneity issues and reverse causation with the contemporaneous CR. Thus, potentially
causing the regression to capture two effects, not only that of CR to HPI, but also HPI to CR.
We remove the contemporaneous CR to attempt to correct for this; the results are reported in
Table 5.1 Panel B. When we remove the contemporaneous CR from the regressions we find
that both HPI and DC coefficients become significant. The immediate and intuitive relationship
between DC and CR continues to hold. The initial positive and counter-intuitive response of
HPI reported in Panel A persists. However, now the negative response definitely outweighs the
positive. The persistence of the initial counter-intuitive response of HPI, accompanied with the
very small p-value suggests that the result is not simply due to endogeneity and reverse
causation.
ย 
	
 ย 
	
 ย 
34
The OLS results presented are purely exploratory, as OLS regressions operate under the
assumption of exogeneity of the independent variable (CR) and possible interaction between
the housing prices and housing output is ignored. It is these constraints that motivate the use
of the SVAR model, which relaxes the assumption of exogeneity and allows for dynamic
interaction between variables. We now move to examine the results of our SVAR model
analysis.
5.2 Structural Vector Auto-Regression Model:
5.2.1 SVAR (I)
We will refer to our base model - shown algebraically in equation (4.3) - as SVAR(I), which
models three primary variables; cash rate, house prices and housing output represented by the
official cash rate (CR), house price index (HPI) and dwelling commencements (DC).
5.2.2 Lag Selection
To ascertain the optimal lag level specification for SVAR(I) we compare the SVAR-lag-
selection-criteria at a maximum of four lags.7
The criteria include the likelihood ratio (LR),
final prediction error (FPE), Akaike information criterion (AIC), Shwartz information criterion
(SIC) and Hannan-Quinn criterion (HQ). Reported in Table 5.2 are the results indicating which
number of lags is recommended as optimal, according to each individual criterion. The results
are mixed, some criteria recommending two lags, others four, although, the majority do in fact
recommend to specify SVAR(I) at two lags. We consider serial correlation in the residuals of
SVAR(I) at two and four lag orders, to resolutely conclude which lag specification is optimal.
Table 5.2	
 ย 
Optimal SVAR Lag Selection Criteria	
 ย 
Lag LogL LR FPE AIC SIC HQ
0 156.7792 NA 9.27e-06 -3.075585 -2.997430 -3.043954
1 394.5925 456.6016 9.54e-08 -7.651851 -7.339231 -7.525328
2 431.3878 68.43909 5.47e-08 -8.207755 -7.660669 -7.986340
3 436.3034 8.848142 5.95e-08 -8.126068 -7.344517 -7.809760
4 447.1006 18.78711 5.75e-08 -8.162012 -7.145995 -7.750812
7
Early experimentation ran up to 12 lags, however no important cases changed as a result of the increased
maximum lag.
ย 
	
 ย 
	
 ย 
35
5.2.3 Residual Serial Correlation LM Test
The null hypothesis of the residual serial correlation test is that no serial correlation is present
in the residuals, shown algebraically below in equation (5.1), where ๐œŒ; represents residual
autocorrelation at lag i. We test for first to fourth order autocorrelation, for both two and four
lags, therefore i = two and four.
Table 5.3 reports the results of our testing of first to fourth-order residual serial correlation at
two and four lags. When specified with two lags, ie i=2, the null hypothesis is rejected at the
fourth lag, suggesting the presence of fourth-order residual serial correlation; which is not
entirely unexpected when using seasonally adjusted quarterly data. However, when tested with
four lags, the fourth-order correlation is removed at the expense of the introduction of first-
order serial correlation. Therefore, a four lag specification is not clearly superior to a two lag
specification. We proceed to report SVAR(I) specified at two lags, in the interest of preserving
simplicity in our model and reducing the degree of statistical noise introduced.
๐ปJ =	
 ย  ๐œŒ; =
๐ถ๐‘œ๐‘ฃ ๐‘ˆ8	
 ย , ๐‘ˆ8>;
๐‘‰๐‘Ž๐‘Ÿ ๐‘ˆ8
= 0
๐ปV = ๐œŒ; =
๐ถ๐‘œ๐‘ฃ(๐‘ˆ8	
 ย , ๐‘ˆ8>;)
๐‘‰๐‘Ž๐‘Ÿ(๐‘ˆ8)
โ‰  0
(5.1)
ย 
	
 ย 
	
 ย 
36
Table 5.3	
 ย 
SVAR Residual Serial Correlation LM Test	
 ย 
Tested at 2 lags
Lags LM-Stat P-Value
1 11.05622 0.2719
2 5.730775 0.7665
3 15.75080 0.0723
4 20.96760 0.0128
Tested at 4 lags
Lags LM-Stat P-Value
1 17.30800 0.0441
2 5.339288 0.8038
3 16.17685 0.0633
4 15.48913 0.0783
Probs from chi-square with 9 df.
5.2.4 Impulse Response Functions
An impulse response function (IRF), graphically traces the responses of the three variables in
SVAR(I) to an increase of each equationโ€™s standard error, which are orthogonal, based on the
Cholesky decomposition. This allows us to observe the response of one variable to a shock or
โ€˜impulseโ€™ in another, and thus study causality and the relationship shared between two variables
in our model.
Figure 5.1 presents all nine IRFs for SVAR(I), one for each variable (CR, HPI, DC) with
respect to the one standard deviation shock to the error of each of the equations, over a 10
period time horizon. The IRFs of particular interest to our analysis are those in the bottom left
corner that trace the response of HPI and DC to a shock in CR. Secondarily, are IRFs positioned
in the top right hand corner, detailing the response of CR to HPI and DC shocks. The IRFs
confirm what the OLS assessment signaled; HPI initially responds counterintuitively to a
contractionary CR shock, increasing upon a CR increase. This behaviour is illustrative of the
broader price puzzle phenomenon that has been documented across various markets, whereby
ย 
	
 ย 
	
 ย 
37
prices and thus inflation increase, in response to contractionary monetary policy. DC responded
as theory would expect, decreasing significantly upon a positive CR shock, although more
immediately, recalling our discussion of the housing market in chapter two, where we
explained the response of housing output to changes in the housing market is understood to be
delayed. We also learn from CR responses to shocks of HPI and DC that CR appears to be
unresponsive to the Australian housing market: not shifting significantly upon either a HPI or
DC shock. In the context of SVAR(I) this suggests that CR is exogenous to the Australian
housing market and that monetary policy doesnโ€™t respond systematically to occurrences in the
housing market.
Less central to our analysis are the IRFs of the own effects of the variables, reported along the
main diagonal, which as we would expect are all positive and significant. The interactions
between HPI and DC are also as predicted, under the intuition of basic demand and supply of
price and output variables: DC decreases upon a positive HPI shock while HPI increases upon
a positive DC shock.
When estimated at four lags, the results of SVAR(I) are comparable to the two lag specification.
However, as expected, volatility - particularly the price puzzle behavior of HPI - is
exacerbated.8
8
See Appendix Figure A.1, for IRFs of SVAR(I) over 10 periods, using a four lag specification.
ย 
	
 ย 
	
 ย 
38
Figure 5.1
Impulse Responses of SVAR(I) over 10 periods
ย 
	
 ย 
	
 ย 
39
5.2.5 Formal Test of Cash Rate Exogeneity
Our OLS analysis and IRFs generated from our SVAR(I) model provide insight into whether
CR is endogenous or in fact operates exogenously to the Australian housing market; both
suggest that CR is exogenous. To formalise our assessment of whether monetary policy is
exogenous within our SVAR(I) model, and thus independent of the Australian housing market,
we execute the block exogeneity Wald test, shown algebraically below in equation (5.2):
The null hypothesis of the test is that the coefficients of the excluded variables - in the case of
SVAR(I), shown again in equation (5.3), these are house price index (HPI) and dwelling
commencements (DC) - are jointly not significantly different to 0. The null hypothesis
therefore concludes that the dependent variable - cash rate (CR) - does not operate in response
to the excluded variables. In sum, the null hypothesis is that CR is exogenous to HPI and DC
when considered jointly; and thus the Australian housing market.
๐ปJ =	
 ย  ๐›ฝ?; =	
 ย  ๐›ฝ@; = 0
	
 ย  ๐ปV =	
 ย  ๐›ฝ?; =	
 ย  ๐›ฝ@; โ‰  0
๐ถ๐‘…8 =	
 ย  ๐›ฝ9 + ๐›ฝ:;
<
;=9
๐ถ๐‘…8>; +	
 ย  ๐›ฝ?;
<
;=9
๐ป๐‘ƒ8>; +	
 ย  ๐›ฝ@;
<
;=9
๐ป๐‘‚8>; +	
 ย  ๐œ–C8	
 ย 
๐ป๐‘ƒ8 =	
 ย  ๐›พ9 + ๐›พ:;
<
;=9
๐ถ๐‘…8>; +	
 ย  ๐›พ?;
<
;=9
๐ป๐‘ƒ8>; +	
 ย  ๐›พ@;
<
;=9
๐ป๐‘‚8>; +	
 ย  ๐œ–E8	
 ย 
๐ป๐‘‚8 =	
 ย  ๐›ฟ9 + ๐›ฟ:;
<
;=9
๐ถ๐‘…8>; +	
 ย  ๐›ฟ?;
<
;=9
๐ป๐‘ƒ8>; +	
 ย  ๐›ฟ@;
<
;=9
๐ป๐‘‚8>; +	
 ย  ๐œ–G8
(5.3)
(5.2)
ย 
	
 ย 
	
 ย 
40
Table 5.4 reports the p-value computed for the Wald Test. Based on the results, we cannot
reject the null hypothesis and conclude in agreement with our OLS and IRF analysis, that CR
is exogenous, and unresponsive to the Australian housing market, in the context of SVAR(I).
By extension, we conclude in SVAR(I), that there exists no evidence of asset price targeting in
Australian monetary policy, with respect to the Australian housing market.
	
 ย 
Table 5.4
SVAR Block Exogeneity Wald Test
Dependent Variable
Cash Rate (CR)
Excluded Variables P-Value
ln(HPIt), ln(DCt) 0.5709
5.3 Sensitivity Analysis:
Upon specification of our base model, SVAR(I) a multitude of decisions, many of which we
do not have firm theoretical ground to guide our choices, were made. Choices regarding the
inclusion of variables, and use of data are examples of decisions that another reasonable
researcher may have specified differently. In our sensitivity analysis we assess the robustness
of the three principal findings of SVAR(I): the price puzzle behavior of HPI in response to CR,
the intuitive response of DC to CR, and the exogeneity of the CR to the Australian housing
market by replacing our base model specifications with sensible alternative assumptions.
We will now outline the sensitivity analysis carried out on the findings of SVAR(I), and draw
attention only to cases where the findings were no longer robust or are particularly interesting.
All sensitivity analysis is reported at the two lag specification, for cases where a lag
specification other than two was optimal we will draw attention to this.
ย 
	
 ย 
	
 ย 
41
5.3.1 ย  Alternative Variable Representation
A. ย  Implicit Price Deflator (IPD) substituted for HPI
B. ย  Gross Fixed Capital Formation (GFCF) substituted for DC
C. ย  Dwelling Approvals (DA) substituted for DC
D. ย  Finance Commitments (FC) substituted for DC
We substitute alternate data sets as proxies for house prices and housing output. We make use
of the following data sets, implicit price deflator (IPD) as representative of house prices,
dwelling approvals (DA), gross fixed capital formation index (GFCF) and finance
commitments (FC), as representative of housing output.
Results of (A), (B) and (C) substantiate the three central findings of SVAR(I).9
The price puzzle
persists and the response of housing output is comparable to SVAR(I) when IPD is substituted
for HPI and again when DA and GFCF are substituted for DC, although the volatility of both
is exacerbated. Further, CR remains unresponsive to the housing market, as the IRFs show
minimal movement, and more formally as we fail to reject the null hypothesis of the block
exoegeneity test.10
Finance Commitments
The results of (D) - where FC are substituted for DC as representative of housing output -
substantiate the first and second main findings of SVAR(I), that of the price puzzle behavior
of house prices and the intuitive response of housing output, although the latter appears
differently to SVAR(I).11
Although from first glance at Figure 5.2, the timing of the response
of FC to a CR shock does not appear comparable to SVAR(I), a similar story is in fact being
told. FC decreases immediately to a positive CR shock, and begins to return to its original level
immediately after that. When we consider FC as the proxy for housing output, we understand
that FC precedes DA, DC and housing investment, explaining the timing differences between
the response of housing output present in D and SVAR(I).
9
The IRFs are reported in the Appendix as Figure A.2, A.3 and A.4, respectively.
10
Results of the block exogeneity Wald test are reported in the Appendix Table A.1.
11
FC measures the numerical value of finance undertaken by those purchasing new homes, both owner
occupiers and investors.
ย 
	
 ย 
	
 ย 
42
The finding of CR exogeneity, however, no longer holds as it is observable in Figure 5.2 that
CR is responsive to a FC shock. This is confirmed by the p-values of the block exogeneity test
- although only just significant - reported in Table 5.5. We reject the null hypothesis of the test,
concluding that under the alternative specifications of (D), CR is responsive to the Australian
housing market. Further, in the Appendix Table A.2, we consider Granger causality and
conclude FC is the Granger cause of CR; we do not find HPI to be the Granger cause of CR.
Interestingly this result is inconsistent to the finding of CR exogeneity in SVAR(I). This
difference potentially suggests that the RBA is responsive to the extent to which people
accumulate debt in the housing market, irrespective of the level of DC or housing output, per
se. Further, as FC occurs before DC, perhaps the RBA reacts to FC, as a predictor of future
housing activity, even though it does not react to DC.
Figure 5.2
Impulse Responses of FC and CR over 10 Periods
	
 ย 
ย 
	
 ย 
	
 ย 
43
Table 5.5
SVAR Block Exogeneity Wald Test
5.3.2 ย  Additional Variables
In their research into the price puzzle in VAR analysis, Rusnak et al (2011) argue its presence
is largely due to model mis-specification, rather than an actual case of the price puzzle. The
paper argues that most cases of the price puzzle can be eliminated or significantly reduced
when the model is correctly specified, particularly when omitted variables are included. In an
effort to reduce omitted variable bias we introduce a range of variables potentially significant
to our results:
A. ย  Building Costs (BC)
B. ย  Disposable Income Growth (YdG)
C. ย  Australian โ€“ Chinese Currency Exchange Rate (ACX)
D. ย  Commodity Price Index (CI)
BC is included to capture the increasing costs of homebuilding, YdG is included to account for
the possibility that disposable income growth may be driving house price increases. Chinese
demand for Australian property is by far the largest foreign demand in the Australian housing
market, as the Foreign Investment Review Board reported Chinese purchasers spent $24.3
billion on Australian property in 2014-2015 alone (Maley 2016). Anecdotally it has been
suggested that this strong Chinese demand for Australian property is driving house price
increases. We include the Australian โ€“ Chinese exchange rate (ACX) as a variable intended to
tease out the effects of Chinese demand. Lastly, in previous literature, including the paper by
Rusnak, Havranek and Horvath (2011) it has been raised that the inclusion of commodity prices
(CI) may alleviate price puzzle behavior from a SVAR model that was mis-specified. The
inclusion of CI stems from Simโ€™s (1992) theory that central banks respond to expectations of
higher future inflation by contracting the cash rate, but not enough to prevent inflation from
Dependent Variable
Cash Rate
Excluded Variables P-Value
ln(HPIt) ln(FCt) 0.0409
ย 
	
 ย 
	
 ย 
44
actually rising, and that the central bank has information about future inflation that is not
captured in the SVAR model. It is understood that CI captures this information.
Initially we introduce these additional variables one at a time at the end of our model, ordered
last, after DC. We recognise the ordering of variables when using a Cholesky decomposition
can be important, and will revisit the order later in our analysis.
Results from (A), (B), (C) and (D) all corroborate the price puzzle behavior of HPI and the
intuitive response of DC to a positive CR shock.12
CR exogeneity is robust for only (A) and
(B); in these cases we fail to reject the null hypothesis of the block exogeneity when estimated
at 2 lags.13
We will deal with the exceptions in more detail. Exogeneity does not hold when we
estimate (D) at two lags.14
Upon assessment of Granger causality, we find CI to be a Granger
cause of CR; HPI and DC are not Granger causes of CR.15
However, when we estimate (C) at
its optimal four lag specification, CR exogeneity holds. We will now focus specifically on the
case of (C) and return to (D) when considering variable ordering.
Australian โ€“ Chinese Exchange Rate
In Figure 5.3 we observe evidence of CR responding to a positive ACX shock; appreciation of
the Australian dollar against the Chinese Yuan. From the p-value reported in Table 5.6 we reject
the null hypothesis of the block exogeneity test for the specification of (C), therefore, the
finding of CR exoegeneity is no longer robust, and CR is deemed endogeneous to theAustralian
housing market. Further, as reported in the Appendix Table A.5, ACX is a Granger cause of
CR. HPI and DC are found to continue to not be Granger causes of CR. CR continues to be
unresponsive to HPI and DC (see Appendix Figure A.8), yet responsive to ACX, suggesting
that the RBA is influenced more by exchange rates in policy than the housing market. This is
quite plausible given the public attention former Governor Glenn Stevens drew to exchange
rates during his time at the RBA.
12
See Appendix Figures A.5 โ€“ A.10. Where 4 lags were optimal (A & D) we present at both two and four lags.
13
See Appendix Table A.3.
14
See Appendix Table A.4.
15
See Appendix Table A.5	
 ย 
ย 
	
 ย 
	
 ย 
45
Table 5.6
SVAR Block Exogeneity Wald Test
Dependent Variable
Cash Rate
Excluded Variables P-Value
ln(HPIt), ln(DCt),
ln(ACXt)
0.0421
Variable Ordering
In using Cholesky decomposition, the ordering of variables in our model is potentially
significant to the results our model yields. We do not alter the ordering of our base variables
(CR, HPI, DC), as the original order, in terms of responsiveness makes intuitive sense. We
reorder the additional variables (A) - (D), from first in our model to third.
The reordering of our additional variables does not alter their effects, the results from these
specifications continue to reinforce the findings of SVAR(I), with the exclusion of (C), as CR
continues to be responsive to ACX and the housing market. The reordering of perhaps the most
interest and potential significance is the ordering of CI first in our model, before the cash rate,
as this specification is in alignment with existing literature that argues the inclusion of CI is
effective in reducing - if not removing - evidence of a price puzzle, as CI, when ordered first
in the SVAR captures the future inflation expectations the central bank is believed to be
Figure 5.3
Impulse Response of CR to ACX over 10 Periods
	
 ย 
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016
Mollie_Urquhart_Dissertation_2016

More Related Content

Viewers also liked

Designshow3
Designshow3Designshow3
Designshow3guestd8ac2a
ย 
Hadoop Presentation - PPT
Hadoop Presentation - PPTHadoop Presentation - PPT
Hadoop Presentation - PPTAnand Pandey
ย 
Iot with-the-best & VSCP
Iot with-the-best & VSCPIot with-the-best & VSCP
Iot with-the-best & VSCPAke Hedman
ย 
SISTEM PGM FI
SISTEM PGM FISISTEM PGM FI
SISTEM PGM FIalohapoint
ย 
Ambient Advertising - Manu Melwin Joy
Ambient Advertising - Manu Melwin JoyAmbient Advertising - Manu Melwin Joy
Ambient Advertising - Manu Melwin Joymanumelwin
ย 
ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™
ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™
ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™sofyanasrun
ย 
4a mdc-planificacin-1-161205221513
4a mdc-planificacin-1-1612052215134a mdc-planificacin-1-161205221513
4a mdc-planificacin-1-161205221513Fausto Yumisaca
ย 

Viewers also liked (9)

Designshow3
Designshow3Designshow3
Designshow3
ย 
Ozono troposfรฉrico
Ozono troposfรฉricoOzono troposfรฉrico
Ozono troposfรฉrico
ย 
Hadoop Presentation - PPT
Hadoop Presentation - PPTHadoop Presentation - PPT
Hadoop Presentation - PPT
ย 
Iot with-the-best & VSCP
Iot with-the-best & VSCPIot with-the-best & VSCP
Iot with-the-best & VSCP
ย 
SISTEM PGM FI
SISTEM PGM FISISTEM PGM FI
SISTEM PGM FI
ย 
Ambient Advertising - Manu Melwin Joy
Ambient Advertising - Manu Melwin JoyAmbient Advertising - Manu Melwin Joy
Ambient Advertising - Manu Melwin Joy
ย 
ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™
ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™
ๆ—ฅๆœฌ่ชž็ทใพใจใ‚ N2 ่ชžๅฝ™
ย 
4a mdc-planificacin-1-161205221513
4a mdc-planificacin-1-1612052215134a mdc-planificacin-1-161205221513
4a mdc-planificacin-1-161205221513
ย 
Innovaciรณn en servicios y mercados de gas natural. Fernando Impuesto
Innovaciรณn en servicios y mercados de gas natural. Fernando ImpuestoInnovaciรณn en servicios y mercados de gas natural. Fernando Impuesto
Innovaciรณn en servicios y mercados de gas natural. Fernando Impuesto
ย 

Similar to Mollie_Urquhart_Dissertation_2016

Computational methods of Hepatitis B virus genotyping
Computational methods of Hepatitis B virus genotypingComputational methods of Hepatitis B virus genotyping
Computational methods of Hepatitis B virus genotypingNguyen Nhat Tien
ย 
PERARES_LK6_Full_paper_book_April_2014
PERARES_LK6_Full_paper_book_April_2014PERARES_LK6_Full_paper_book_April_2014
PERARES_LK6_Full_paper_book_April_2014Diana Saplacan
ย 
MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...
MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...
MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...Alejandro Freund
ย 
Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...
Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...
Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...Manos Tsimperis
ย 
Financial modelling with forward looking information.pdf
Financial modelling with forward looking information.pdfFinancial modelling with forward looking information.pdf
Financial modelling with forward looking information.pdfFinanzasEmpresariale
ย 
final dissertation
final dissertationfinal dissertation
final dissertationAbubakar Musa
ย 
Assessing The Impact Of Academic Literacy Interventions In Higher Education ...
Assessing The Impact Of Academic Literacy Interventions In Higher Education  ...Assessing The Impact Of Academic Literacy Interventions In Higher Education  ...
Assessing The Impact Of Academic Literacy Interventions In Higher Education ...Leonard Goudy
ย 
Dessertation for Bechalor in Business Administration
Dessertation for Bechalor in Business AdministrationDessertation for Bechalor in Business Administration
Dessertation for Bechalor in Business AdministrationOjok Francis
ย 
IJLTER.ORG Vol 19 No 12 December 2020
IJLTER.ORG Vol 19 No 12 December 2020IJLTER.ORG Vol 19 No 12 December 2020
IJLTER.ORG Vol 19 No 12 December 2020ijlterorg
ย 
An Investigation into Perceptions of Vietnamese Teachers on Models of English...
An Investigation into Perceptions of Vietnamese Teachers on Models of English...An Investigation into Perceptions of Vietnamese Teachers on Models of English...
An Investigation into Perceptions of Vietnamese Teachers on Models of English...Marcus Martins
ย 
Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...
Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...
Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...Gwen Naylor
ย 
Katlego_Pule_674426_Research_report_final_submission
Katlego_Pule_674426_Research_report_final_submissionKatlego_Pule_674426_Research_report_final_submission
Katlego_Pule_674426_Research_report_final_submissionKatlego Pule
ย 
Undergraduate Dissertation- Timothy McKeown
Undergraduate Dissertation- Timothy McKeownUndergraduate Dissertation- Timothy McKeown
Undergraduate Dissertation- Timothy McKeownTim McKeown
ย 
Vol 8 No 1 - October 2014
Vol 8 No 1 - October 2014Vol 8 No 1 - October 2014
Vol 8 No 1 - October 2014ijlterorg
ย 
Accounting And Financial Statements
Accounting And Financial StatementsAccounting And Financial Statements
Accounting And Financial StatementsMelinda Watson
ย 
lapin-ci
lapin-cilapin-ci
lapin-ciCarly Lapin
ย 
Effectiveness of authentic materials on extensive reading in developing stude...
Effectiveness of authentic materials on extensive reading in developing stude...Effectiveness of authentic materials on extensive reading in developing stude...
Effectiveness of authentic materials on extensive reading in developing stude...NuioKila
ย 

Similar to Mollie_Urquhart_Dissertation_2016 (20)

Computational methods of Hepatitis B virus genotyping
Computational methods of Hepatitis B virus genotypingComputational methods of Hepatitis B virus genotyping
Computational methods of Hepatitis B virus genotyping
ย 
Innovation Strategies in Tourism industry
Innovation Strategies in Tourism industryInnovation Strategies in Tourism industry
Innovation Strategies in Tourism industry
ย 
PERARES_LK6_Full_paper_book_April_2014
PERARES_LK6_Full_paper_book_April_2014PERARES_LK6_Full_paper_book_April_2014
PERARES_LK6_Full_paper_book_April_2014
ย 
MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...
MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...
MSc. Thesis - Alejandro Freund - A Proposal for Sustainable Development in Ru...
ย 
Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...
Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...
Master's Thesis_Identification of Investment Opportunities in Urban Regenerat...
ย 
Financial modelling with forward looking information.pdf
Financial modelling with forward looking information.pdfFinancial modelling with forward looking information.pdf
Financial modelling with forward looking information.pdf
ย 
final dissertation
final dissertationfinal dissertation
final dissertation
ย 
Assessing The Impact Of Academic Literacy Interventions In Higher Education ...
Assessing The Impact Of Academic Literacy Interventions In Higher Education  ...Assessing The Impact Of Academic Literacy Interventions In Higher Education  ...
Assessing The Impact Of Academic Literacy Interventions In Higher Education ...
ย 
Dessertation for Bechalor in Business Administration
Dessertation for Bechalor in Business AdministrationDessertation for Bechalor in Business Administration
Dessertation for Bechalor in Business Administration
ย 
IJLTER.ORG Vol 19 No 12 December 2020
IJLTER.ORG Vol 19 No 12 December 2020IJLTER.ORG Vol 19 No 12 December 2020
IJLTER.ORG Vol 19 No 12 December 2020
ย 
An Investigation into Perceptions of Vietnamese Teachers on Models of English...
An Investigation into Perceptions of Vietnamese Teachers on Models of English...An Investigation into Perceptions of Vietnamese Teachers on Models of English...
An Investigation into Perceptions of Vietnamese Teachers on Models of English...
ย 
EN3100 FINAL
EN3100 FINALEN3100 FINAL
EN3100 FINAL
ย 
Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...
Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...
Paupers, Vagabonds and Oliver Twist An Investigation of Public Perceptions of...
ย 
Katlego_Pule_674426_Research_report_final_submission
Katlego_Pule_674426_Research_report_final_submissionKatlego_Pule_674426_Research_report_final_submission
Katlego_Pule_674426_Research_report_final_submission
ย 
Undergraduate Dissertation- Timothy McKeown
Undergraduate Dissertation- Timothy McKeownUndergraduate Dissertation- Timothy McKeown
Undergraduate Dissertation- Timothy McKeown
ย 
Full Desso
Full DessoFull Desso
Full Desso
ย 
Vol 8 No 1 - October 2014
Vol 8 No 1 - October 2014Vol 8 No 1 - October 2014
Vol 8 No 1 - October 2014
ย 
Accounting And Financial Statements
Accounting And Financial StatementsAccounting And Financial Statements
Accounting And Financial Statements
ย 
lapin-ci
lapin-cilapin-ci
lapin-ci
ย 
Effectiveness of authentic materials on extensive reading in developing stude...
Effectiveness of authentic materials on extensive reading in developing stude...Effectiveness of authentic materials on extensive reading in developing stude...
Effectiveness of authentic materials on extensive reading in developing stude...
ย 

Mollie_Urquhart_Dissertation_2016

  • 1. ย  An Empirical Assessment of the Role of the Australian Housing Market in the Transmission of Monetary Policy Mollie Annie Urquhart Economics Honours Dissertation Submitted in partial fulfilment of the requirements for the degree of Bachelor of Commerce (Honours) Supervised by Professor Nicolaas Groenewold Submission date: 14 November 2016
  • 2. ย  ย  i Abstract Since the early 1990s the Reserve Bank of Australia (RBA) has explicitly targeted a medium term level of inflation of between two and three per cent. As housing costs account for approximately 22 per cent of the Consumer Price Index basket โ€“ the RBAโ€™s primary measure of inflation โ€“ and housing investment four per cent of Gross Domestic Product, it is clear the Australian housing market is intrinsically important to the RBAโ€™s objective; making an assessment of the role of the housing market in monetary policy transmission an important exercise. This dissertation pursues this objective using data for the period 1990Q1 to 2015Q4 and a structural vector autoregression model to simulate the interaction of house prices, housing investment and the cash rate in response to a monetary policy shock. Results indicate that (1) house prices initially react positively to a monetary contraction, suggesting a form of the โ€˜price puzzleโ€™ is operational in the housing market, (2) the quantity of housing investment falls in response to a contractionary monetary shock and (3) the cash rate operates independently of the housing market. Alternative data sets are used, additional variables are included and sub-periods are created to determine the robustness of the results; reducing the likelihood of model mis-specification. The results are remarkably robust.
  • 3. ย  ย  ย  ii Acknowledgements ย  First and foremost, I would like to thank my supervisor Prof. Nicolaas Groenewold; without whom this dissertation would not have been possible. Thank you Nic for your guidance and support throughout this year; I am grateful for how pleasant you have made the experience. Thank you Elisa Birch, Anu Rammohan and particularly Leandro Magnusson for your continued guidance. Thank you also to my lecturers Peter Robertson, Rodney Tyers, Michael McLure and Yanrui Wu. I would also like to thank my parents Mr David Urquhart and Mrs Wendy Urquhart. The sacrifices they have made and the support they have provided throughout my entire education has meant the world. This dissertation is a testament to their belief in my ability. Thank you Dean van Kwawegen for proofreading this dissertation. Finally, thank you to the 2016 Economics Honours cohort, it has been a pleasure to write my dissertation alongside you all.
  • 4. ย  ย  ย  iii Declaration Unless otherwise acknowledged in the text or acknowledgements, the work presented in this dissertation is my own original work. This dissertation has 14,684 words, including appendices. I give permission for the Economic Discipline to use my honours dissertation as an example of a dissertation which may get distributed to staff and future students.โ€จ I do not give permission for the Economic Discipline to use my honours dissertation as an example of a dissertation which may get distributed to staff and future students. ___________________________________ _______________________ Mollie Annie Urquhart Date 14th November 2016
  • 5. ย  ย  ย  iv Table of Contents Abstract......................................................................................................................................i Acknowledgements ................................................................................................................. ii Declaration.............................................................................................................................. iii Table of Contents....................................................................................................................iv List of Tables ...........................................................................................................................vi List of Figures........................................................................................................................ vii Chapter One โ€“ Introduction ...................................................................................................1 Chapter Two โ€“ Australian Monetary Policy and the Housing Market ..............................4 2.1 Monetary Policy, Then and Now............................................................................................... 4 2.2 Monetary Policy Transmission Channels................................................................................. 5 2.3 The Australian Housing Market ............................................................................................... 7 2.4 Concluding Remarks.................................................................................................................. 9 Chapter Three โ€“ Literature Review.....................................................................................10 3.1 Monetary Policy Literature..................................................................................................... 10 3.2 International Housing Market Literature.............................................................................. 12 3.3 Australian Housing Market Literature.................................................................................. 13 3.4 Asset Price Targeting Literature ............................................................................................ 14 3.5 Price Puzzle Literature ............................................................................................................ 15 3.6 Contribution to the Existing Literature ................................................................................. 16 Chapter Four โ€“ Method and Data .......................................................................................18 4.1 Ordinary Least Squares Method (OLS)................................................................................. 18 4.2 Structural Vector Autoregressive Model (SVAR)................................................................. 19 4.2.1 Identification of Monetary Policy Shock ............................................................................ 20 4.2.2 Impulse Response Functions............................................................................................... 20 4.2.3 Block Exogeneity Wald Test .............................................................................................. 21 4.3 Variables.................................................................................................................................... 21 4.4 Data............................................................................................................................................ 22 4.4.1 Stationarity .......................................................................................................................... 23 4.5 Primary Variable Tests............................................................................................................ 25
  • 6. ย  ย  ย  v 4.6 Robustness Variables Tests ..................................................................................................... 28 4.6.1 Alternative Data Sets........................................................................................................... 28 4.6.2 Additional Variables ........................................................................................................... 30 Chapter Five - Results ...........................................................................................................32 5.1 Preliminary Analysis................................................................................................................ 32 5.1.1 Ordinary Least Square Regression Analysis....................................................................... 32 5.2 Structural Vector Auto-Regression Model ............................................................................ 34 5.2.1 SVAR (I) ............................................................................................................................. 34 5.2.2 Lag Selection....................................................................................................................... 34 5.2.3 Residual Serial Correlation LM Test .................................................................................. 35 5.2.4 Impulse Response Functions............................................................................................... 36 5.2.5 Formal Test of Cash Rate Exogeneity ................................................................................ 39 5.3 Sensitivity Analysis................................................................................................................... 40 5.3.1 Alternative Variable Representation............................................................................... 41 5.3.2 Additional Variables ....................................................................................................... 43 5.3.3 Sub-Periods ..................................................................................................................... 46 5.3.4 Additional Sensitivity Testing ........................................................................................ 49 5.4 Concluding Remarks................................................................................................................ 52 Chapter Six - Conclusion ......................................................................................................54 References...............................................................................................................................56 Appendix.................................................................................................................................64
  • 7. ย  ย  ย  vi List of Tables Table 4.1 Unit Root Tests.......................................................................................................26 Table 4.2 Breusch-Godfrey Serial Correlation LM Test ........................................................27 Table 4.3 Unit Root Tests.......................................................................................................28 Table 4.4 Breusch-Godfrey Serial Correlation LM Test ........................................................30 Table 4.5 Unit Root Tests.......................................................................................................31 Table 4.6 Breusch-Godfrey Serial Correlation LM Test ........................................................31 Table 5.1 OLS Regression Results ..........................................................................................33 Table 5.2 Optimal VAR Lag Selection Criteria.......................................................................34 Table 5.3 SVAR Residual Serial Correlation LM Test ...........................................................36 Table 5.4 SVAR Block Exogeneity Wald Test .......................................................................40 Table 5.5 SVAR Block Exogeneity Wald Test .......................................................................43 Table 5.6 SVAR Block Exogeneity Wald Test .......................................................................45 Table 5.7 SVAR Block Exogeneity Wald Test .......................................................................48
  • 8. ย  ย  ย  vii List of Figures Figure 1.1 Australian Housing Lending Rates...........................................................................2 Figure 1.2 Housing Prices and Interest Rates............................................................................2 Figure 2.1 Australian Inflation...................................................................................................5 Figure 2.2 Response of the Australian Housing Market to Expansionary Monetary Policy.....9 Figure 4.1 Cash Rate, ln(House Price Index), ln(Dwelling Commencements) Plots..............24 Figure 5.1 Impulse Responses of SVAR(I) over 10 Periods ...................................................38 Figure 5.2 Impulse Responses of FC and CR over 10 Periods................................................42 Figure 5.3 Impulse Response of CR to ACX over 10 Periods.................................................45 Figure 5.4 Impulse Response of HPI to CR over 10 Periods...................................................47 Figure 5.5 Impulse Response of DC to CR over 10 Periods ...................................................48 Figure 5.6 Impulse Response of CR and MR over 10 Periods................................................50 Figure 5.7 Impulse Responses of CR and HPI (Cumulative) over 30 Periods........................51 Figure 5.8 Accumulated Impulse Response of DC (First Differenced) over 10 Periods ........52
  • 9. ย  ย  ย  1 Chapter One โ€“ Introduction Since the early 1990s the Reserve Bank of Australia (RBA) has explicitly targeted inflation levels of between two and three per cent over the medium term. This target is met through the exercising of monetary policy โ€“ specifically changing the cash rate โ€“ to influence economy wide interest rates and ultimately manage economic conditions. Costs of housing investment borne by owner occupiers account for approximately nine per cent of the Consumer Price Index (CPI) basket โ€“ the primary measure of annual inflation โ€“ while housing costs in sum compose 22.3 per cent (Reserve Bank of Australia 2014).1 Together Figure 1.1 and 1.2 detail the strong pass-through of cash rate changes to housing lending rates and the transmission of these effects to house prices. A strong positive relationship is observable between the cash rate and mortgage rate, and a strong negative relationship is evident between the mortgage rate and house prices. As the cash rate holds a strong relationship with house prices, through the mortgage rate, and house prices significantly contribute to inflation, through the CPI, it appears that the transmission of monetary policy to house prices is important to the maintenance of the inflation target. It is surprising, in light of the importance of Australian housing to CPI inflation measurement, and the strong relationship between the cash rate, mortgage rate and house prices, that an assessment of the Australian housing marketโ€™s role in the maintenance of the inflation target has not previously been conducted. 1 The housing component of the Consumer Price Index is calculated by the Australian Bureau of Statistics (ABS) using costs of rent, new dwellings purchased by owner occupiers, maintenance and repair costs, property rates and charges, utilities and household products. In aggregate the housing component accounts for 22.3 per cent of the CPI basket. While rent costs account for 6.71 per cent CPI and new dwelling owner occupier costs account for 8.67 per cent of the CPI.
  • 10. ย  ย  ย  2 Our research is interested in the transmission of monetary policy shocks to inflation and within this transmission system we are specifically concerned with the role of the housing market. We analyse the effects of changes in the cash rate on house prices and housing investment; which are, respectively, transmitted to inflation directly, through CPI measurement and indirectly, through aggregate demand. Observable in Figure 1.1 and 1.2 is the direct transmission of monetary policy to inflation through house prices. Additionally, an indirect effect upon inflation, transmitted through housing investment is anticipated to be operational. This is a consequence of the intrinsic importance of housing investment to the Australian economy, accounting for approximately four per cent of Gross Domestic Product (GDP) (Australian Industry Group 2015), while also tending to be the most interest rate sensitive component (Berger - Thomson & Ellis 2014), and thus a strong driving force of aggregate demand and ultimately inflation. The relationship between monetary policy and the housing market is noteworthy not only because of these stylised facts regarding our nationโ€™s largest asset class, which rose to six trillion AUD in June 2016, making it three and a half times larger than the Australian share market (Bleby & Greber 2016), but also recently has become increasingly topical, as speculation builds that exponentially increasing housing demand, particularly in Sydney and Melbourne is creating a bubble in the Australian housing market. In light of this it becomes important to assess the role of the Australian housing market in monetary policy transmission, Figure 1.1 Australian Housing Lending Rates (Average interest rate on variable-rate loans) Figure 1.2 Housing Prices and Interest Rates (Reserve Bank of Australia 2016) (Reserve Bank of Australia 2014)
  • 11. ย  ย  ย  3 and also examine whether this has changed over the course of the last 25 years; former RBA Governor Glenn Stevens flagged concern that the ability of monetary policy to influence housing markets is today a channel that โ€œmay not be quite as effective as it once wasโ€ (Stevens 2015). Our work is further motivated by the absence in the existing literature of an Australian analysis that considers the two possible components of the housing market transmission mechanism separately, house prices and housing investment, rather than focusing solely on house prices. Our research is centred in analysing the effects of changes in the cash rate on house prices and housing output; assessing the role of the Australian housing market as a channel of monetary policy transmission. I conjecture the house price mechanism will behave as an immediate effect, as housing stock remains fixed in the short term; while over the longer term the investment channel will become more effective. Additionally, our analysis also addresses the possibility of reverse feedback from both house prices and housing investment, to monetary policy. Examining the housing market as a transmission mechanism of monetary policy offers potentially important policy implications. An understanding of this transmission mechanism in an environment where its effectiveness is a concern of policy makers may allow the RBA to more suitably conduct future monetary policy. The results indicate the investment transmission mechanism is effective in transmitting cash rate changes through the housing market. The house price channel does not operate as intuitively expected; instead, house prices increase as a result of an increase in the cash rate. We argue this is a particular instance of the generally observed โ€˜price puzzleโ€™ phenomenon. The cash rate is found to operate exogenously from house prices and housing investment, indicating the RBA does not set policy in direct response to the Australian housing market. The remainder of this paper is structured as follows. Chapter two explores Australian monetary policy and its transmission through the housing market. Chapter three provides a review of the literature on monetary policy and housing markets. Chapter four describes the econometric methods and data instrumental to our research. Chapter five presents our results. Lastly, chapter six considers the policy implications of our findings.
  • 12. ย  ย  ย  4 Chapter Two โ€“ Australian Monetary Policy and the Housing Market To adequately assess the role of the Australian housing market, to the transmission of monetary policy, a strong grasp of the operation of monetary policy, particularly in the Australian inflation targeting context, and the nature of the Australian housing market is essential. 2.1 Monetary Policy, Then and Now: The nature of current Australian monetary policy is a far cry from its original format as a central banking function of the Commonwealth Bank. Since the Reserve Bank Act 1959 monetary policy in Australia has been operated by a separate central bank authority, the RBA. A flavour of the nature of monetary policy during this early period may be derived from Dr H.C โ€˜Nuggetโ€™ Coombs - Governor of the Commonwealth Bank (1949-1959) and RBA (1960-1968) - endorsement of the view that, โ€œA central bank should be like a good wife. It should manage its household competently and quietly; it should stand ready to assist and advise; it can properly persuade and cajole and on occasions even nag; but in the end it should recognise that the government is the boss.โ€ (Coombs 1971 p63) From this declaration comes the view that monetary policy was a tool to aid fiscal policy, rather than operate within its own framework to achieve separate objectives. During the 1970s period of stagflation, whereby stagnant economic conditions and high inflation were jointly present, the Phillips curve trade off between unemployment and inflation disappeared, and the nature of policy operation shifted greatly. Stephen Grenville declares it is from this period of disruption that monetary policy was first appreciated as a โ€œseparate specialised function, with
  • 13. ย  ย  ย  5 a comparative advantage clearly distinguishable from fiscal policyโ€ (Grenville 2001), necessitating a nominal anchor and framework to constrain the decision-making process of monetary policy. Thus, the stage was set for the abandonment of money targeting, and the adoption of inflation targeting, the framework under which monetary policy currently operates. In the early 1990s the monetary policy objective of price stability was outlined by then RBA Governor Bernie Fraser, as an average rate of inflation between two and three per cent over the medium term, this explicit objective of the inflation targeting framework remains today. Figure 2.1 details the Australian inflation rate as measured by the proportional change in the Consumer Price Index (CPI), extending from the late 1950s to the present day; illustrating that since the introduction of the inflation targeting regime the proportional change in the CPI has in fact remained on average within the two and three per cent band. Figure 2.1 Australian Inflation (Reserve Bank of Australia 2016) 2.2 Monetary Policy Transmission Channels: The IS-LM model is the standard textbook model commonly used for monetary policy analysis, in practise this model captures only one channel of monetary policy transmission; the interest rate channel. In reality the transmission of monetary policy is far more complicated than the operation of the interest rate transmission mechanism alone. The transmission channel literature commonly identifies three further channels, the exchange rate channel, credit channel and asset prices channel.
  • 14. ย  ย  ย  6 The four widely accepted transmission channels of monetary policy are discussed in detail by Mishkin (1995). The interest rate channel, which is commonly demonstrated using the IS-LM model, is regarded as the key mechanism in the basic Keynesian model, traditionally recognised as primarily operating through business decisions about investment spending and later recognised as operational through consumer decisions about housing and durable expenditure. Taylor (1995) argues the importance of the interest rate channel in transmitting effects of monetary policy to the economy, demonstrating that a contractionary policy shock raises the short term nominal interest rate. The higher nominal interest rates therefore lead to a decline in business fixed investment, residential housing investment, consumer expenditure on durables and inventory investment, causing a decline in aggregate output and inflation. The exchange rate transmission channel has, in the view of Mishkin (1995) become more important to the understanding of monetary policy as globalisation increases. This channel operates through the ability of the exchange rate to drive net exports. An increase in the domestic interest rate increases the attractiveness of domestic deposits to foreign currency, causing an appreciation of the domestic currency. By extension an appreciation of domestic currency results in domestic goods being more expensive than foreign goods, thereby causing a fall in net exports, aggregate output and inflation. Obstfeld and Rogoff (1995) emphasise the importance of the exchange rate channel of monetary policy transmission and, in agreement with Taylor (1995), noted that the framework for conducting monetary policy must be international in scope, hence the consideration of the exchange rate transmission mechanism. Bernanke and Gertler (1995) studied the credit channel of monetary policy transmission, illustrating that two effects operate within the credit market to transmit monetary policy effects to the economy, the bank lending effect and balance-sheet effect. The bank lending effect operates upon a contractionary monetary policy shock, as the interest rate charged in the over night money market increases, causing bank reserves to decrease, constraining lending and thus reducing investment, and by extension aggregate output and inflation. The balance-sheet transmission mechanism functions through the net worth of business firms. Contractionary monetary policy causes prices to reduce, thus reducing the net worth of firms, resulting in lenders having less collateral for their loans, reducing lending, and similar to the bank lending effect, investment, aggregate output and inflation decreases
  • 15. ย  ย  ย  7 Lastly, the asset prices mechanism of monetary policy transmission. Meltzer (1995) highlighted monetaristโ€™s objection to the Keynes framework for analysing monetary policy effects on the economy as the focus is on only one asset price, the interest rate. Tobinโ€™s (1969) q theory details the mechanism through which monetary policy affects the economy through the value of assets. When monetary policy is contracted the public have restricted access to money and reduce spending, decreasing the prices of assets such as equities on the stock market. Lower equity prices under the q theory lead to a lower โ€˜qโ€™ value2 - the term representative of the market value of firms divided by the replacement cost of capital - reducing investment spending as it is expensive relative to equity value, which ultimately reduces aggregate output and inflation. Alternatively, Modigliani (1971) argues the asset price mechanism operates through wealth effects on consumption. Under this rationale when asset prices fall the value of wealth decreases causing consumers to reduce consumption, and ultimately as a result aggregate output and inflation declines. Beyond the literature identifying these fundamental channels of transmission exists a vast body of study regarding the effectiveness of monetary transmission to the real economy (Bernanke et al 2005), cross country analysis (Musso 2011) and the relative importance of particular traditional channels (Ibarra 2016). 2.3 The Australian Housing Market: Having detailed the history of Australiaโ€™s inflation targeting monetary policy framework, and the four common transmission channels, it may be acknowledged going forward, the importance of specific asset prices, not only general inflation levels in maintaining economic stability. While asset prices already serve a role in the inflation targeting framework, through the CPI measure of inflation, experiences of many economies, most prominently perhaps, the Unites States in 2008, and the enormous disruption asset bubbles cause (Miller & Stiglitz 2010) has motivated many to consider whether monetary policy should do more to directly influence asset prices. This conjecture leads us to a discussion of one of the most important asset markets to the Australian economy, the housing market; an appreciation of which is fundamental to 2 If q is high the market value of firms is high relative to the replacement cost of its capital, new capital is cheap relative to the market value of business. Companies may issue equity and receive a high price relative to the capital purchased (Mishkin 1995).
  • 16. ย  ย  ย  8 understanding our assessment of its role in transmitting monetary policy throughout the economy. The Australian housing market is characterised by given housing stock in the short term and a heightened sensitivity to interest rates, due to reliance upon borrowed funds. As a consequence of lagged debt approval, land release, building permission and the generally long term nature of housing production, supply response to monetary policy shocks is a delayed and long term process. Due to these characteristics our analysis of the operation of the housing market in response to monetary policy in the short term details a strikingly different story to a long term analysis. Figure 2.2 illustrates the response we expect of the housing market to an expansionary monetary policy shock. On the vertical axis is house prices (PH), the horizontal axis measures housing output (QH). A downward sloping curve (HD ) represents housing demand and a vertical curve represents short term housing supply (HS S.R). In the long run housing supply is gradually more elastic and represented by an upward sloping curve (HS L.R), as the housing supply curve pivots clockwise over the time horizon. The model of the housing market features housing prices and quantity as endogenous variables, the values of which are determined by the intersection of housing demand and housing supply (short run or long run) curves. In the short run housing quantity is determined by the vertical (inelastic) supply curve and is therefore verifiably exogenous. Output becomes truly endogenous in the long run; a determinant of the intersection of housing demand and long run housing supply curves, as supply is elastic and responsive to house prices. In the event of an expansionary policy shock demand for housing increases from HD to HD1 , in the short run as housing stock is fixed, prices in the market increase from HP0 t HP1 , while output remains stable at HQ0 . In the longer term the housing supply curve gradually becomes more elastic, as producers in the market are able to respond to the conditions of the market. In the long term the same expansionary shock results in the price level only rising to HP* , and output increasing to HQ* , as supply is capable of responding to increased demand due to lower interest rates, thus reducing the pressure on the price level in the housing market.
  • 17. ย  ย  ย  9 Figure 2.2 Response of the Australian Housing Market to Expansionary Monetary Policy Unlike most goods markets, the housing market is not characterised in the short term by sticky prices, and output is given in the short term. Thus, we conjecture that in the short run an expansionary monetary policy shock will increase house prices, feeding through to increase inflation. In the long run output is also affected, the direct price effect on inflation remains positive, also operational is an indirect investment effect increasing inflation, aggregate demand. Furthermore, we conjecture that the scale of these effects are relatively larger and that the Australian housing market is more responsive to monetary policy shocks, than the average goods market, due to the increased reliance upon borrowed funds, the majority of which are variable rate mortgages. 2.4 Concluding Remarks: Through discussion of the inflation targeting framework, a review of the transmission channels and an analysis of monetary policy operation within the uniquely characterised Australian housing market, we aim to have illustrated the nature of Australian monetary policy and the housing market, through which policy effects inflation. With this general understanding, an appreciation for the coming literature review regarding monetary policy and the housing market may be formed. Furthermore, an elevated appreciation for the results of our examination of the role of the Australian housing market, to the transmission of monetary policy may be achieved.
  • 18. ย  ย  ย  10 Chapter Three โ€“ Literature Review Recalling chapter one, our preoccupation is the effects of monetary policy upon house prices and housing output, as we assess the role of the housing market in monetary policy transmission, identifying the direct effect of house prices and the indirect effect of housing investment (output) upon inflation. To this end we review literature identifying the effects of monetary policy on output and inflation. A subset of this literature is empirical research, both international and Australian, that analyses the response of the housing market to monetary policy and we survey this specialisation. The principal monetary policy transmission channels outlined in the literature have been surveyed previously in chapter two. Discussion regarding the importance of asset prices to monetary policy has increased, particularly since the Global Financial Crisis and recently, within Australia, due to the housing affordability problem in certain capital cities. In light of this, and the possibility of reverse feedback from the housing market to monetary policy, we review the asset price targeting literature. Lastly, as a precursor to out results, we review literature identifying the price puzzle phenomenon. Common empirical techniques used to assess the transmission of policy shocks to output and price levels will be reviewed in the following chapter regarding modelling and data. 3.1 Monetary Policy Literature: There exists a long empirical literature analysing how exogenous monetary policy shocks affect output and inflation variables. This literature is classified under Monetarism, a school of thought which asserts fairly definite predictions of monetary policy; that the money supply has the ability to influence output and price levels, the foundation for empirical analysis from the likes of Friedman and Fisher regarding monetary policy. Monetarists such as Friedman (1968) theorise that irrespective of a central bankโ€™s targeting framework, monetary policy cannot influence real variables such as unemployment and output growth over the long term, as shocks only affect nominal variables, ie. the price level (inflation), income and interest rates, as per
  • 19. ย  ย  ย  11 the โ€˜liquidity effectโ€™ over an extended time horizon. Daniel (1981) is an example of early empirical literature - extending upon Lucas (1972) - illustrating that announced contractionary monetary policy in a small open economy holds a negative relationship with output, in the short term. This result has become the general consensus view in empirical analysis, as the findings of Sims (1980), Bernanke and Blinder (1992), Eichenbaum (1992) and Leeper and Gordon (1992) all concur that an increase in the main tool of an inflation targeting central bank - the policy interest rate - leads to a decline in output in the short term. Shifting focus, we review literature formalising the effects of monetary policy upon inflation as illustrated in our housing market model in chapter two. Irving Fisher in The Purchasing Power of Money (1911) developed the relationship between changes in monetary policy and changes in general price levels. Since this seminal paper, the literature has become vast. Haque (1985) is representative of early literature, building upon the โ€˜Friedman-typeโ€™ beliefs held by monetarists that monetary policy will not permanently influence real variables, although it would exert an influence over inflation rates, especially in the long run. Haque illustrates with contractionary monetary policy, steady state inflation rate decreases and inflation during the transition path also falls. Furthermore, Karim et al (2011) and De Waal and Van Eyden (2014), are examples of the modern literature examining the relationship between monetary policy and inflation. Karim et al (2011) examines the price effects of monetary policy in the small open economy of New Zealand; where contractionary monetary policy reduced inflation. De Waal and Van Eyden (2014) conduct similar yet extended analysis on the South African economy, investigating the impact of an increase in the official interest rate on both output and inflation, noting Mishkinโ€™s (1995) reference to these elements as the โ€˜timing and effectโ€™ of monetary policy on the economy. Interestingly, the response of inflation suggests a lag of approximately 24 months and evidence of the price puzzle โ€“ the positive relationship between tightened monetary policy and inflation - observable in the first quarter, after which inflation declines. Evident through this brief review of literature examining relationships between monetary policy, output and inflation, is the consensus view that monetary contraction reduces inflation and output. The consensus formalises our previous discussion of monetary policy operation; the transmission channels reviewed in chapter two detail the means through which both effects materialise.
  • 20. ย  ย  ย  12 3.2 International Housing Market Literature: The means through which monetary policy affects output and inflation, are acknowledged as transmission mechanisms of monetary policy; of which the most widely recognised have been discussed at length in the previous chapter. The extension of this literature of greatest interest to this dissertation, is the consideration of the importance of the housing market to the transmission of monetary policy. Internationally, research on the housing market as a transmission channel of monetary policy has been carried out most densely in the United States (US) (Vargas-Silva 2008, Dore et al 2013, Christidou and Konstantinou 2011), while the United Kingdom (UK) (Elbourne 2005), China (Wei et al 2014), South Africa (Gupta et al 2009) and developing economies such as Turkey (Guler 2012) have also considered the housing marketโ€™s role in the transmission of monetary policy to the economy, affecting price levels and output. Empirical analysis of the importance of the housing market as a transmission mechanism of monetary policy in the US has produced largely consistent results. Vargas-Silva (2008) findings suggest that contractionary monetary policy shocks resulted in significantly negative effects on housing starts and residential investment. The vector autoregressive (VAR) model estimated by Christidou and Konstantinou (2011) over the period 1988-2009 similarly assessed the effect of a monetary policy shock on US house prices and US housing investment. Again contractionary monetary policy was found to lead to a reduction in housing investment. Further, this effect in many states is found to be short lived and reversed within less than five years; consistent with the theories of monetarists, that monetary policy is incapable of affecting real variables in the long term. Further, a pronounced and long-lasting fall in housing prices is documented. Analysis using an eight variable structural vector autoregressive (SVAR) model of the UK executed by Elbourne (2005) again illustrated that a contractionary monetary policy shock, resulted in decreased house prices. Implementation of similar econometric analysis to South Africa and Turkey uncovered results that further validate the US and UK analysis. Gupta et al (2009) finds South African house prices increase in response to an expansionary policy shock, in consensus with the results of previous literature. Empirical analysis of Turkey contributes to the consensus view that โ€œthe
  • 21. ย  ย  ย  13 interest rate affects the housing market considerably and house prices, particularly, play an important role in the monetary transmissionโ€. (Guler 2012) The intuitive relationship between monetary policy and house prices held before 2001; this relationship, however, is then undermined, where results illustrate a comparable shock decreases house prices. The response of housing investment is consistently compliant with expectations for the full period; decreasing as monetary policy contracts. Wei et al (2014) executed comparable analysis for the Chinese case and uncovered interesting results that largely contradict the existing consensus; a contractionary monetary policy shock appears to induce positive effects on both housing investment and house prices. The response of prices is consistent with the price puzzle phenomenon; the literature upon which we will expand upon later in our review. 3.3 Australian Housing Market Literature: Domestically, fewer studies have examined the housing market and monetary policy, the majority of which having been carried out over the last decade, as the GFC and the Australian housing affordability problem, ignited the debate over whether monetary policy should respond to the housing market, specifically house prices, see (Mishkin 2007). Within the literature, as flagged by Wadud et al (2012), is a focus on identifying effects of monetary policy on house prices and causes of overvaluation (see Fry et al, 2010), rather than an assessment of the marketโ€™s importance as a transmission channel of monetary policy. Wadud et al (2012) models the impact of monetary policy shocks upon the Australian housing market. A nine variable SVAR model is estimated and impulse responses are generated for the following variables; housing material costs, house prices, GDP, nominal domestic interest rate, inflation rate, foreign interest rate, government spending on housing and the nominal exchange rate against the dollar variables.3 The data, which includes quarterly housing approvals as a proxy for housing investment and the Australian Bureau of Statistics (ABS) house price index as a proxy for house prices, extends from 1974 - 2008.Wadud et al found that the short term interest rate and inflation rate are the main determinants of house prices and house prices significantly affect housing investment. Furthermore, the impulse responses, suggest a 3 The specification of the structural vector autoregression (SVAR) will be outlined in chapter four.
  • 22. ย  ย  ย  14 contractionary monetary policy shock has an immediate positive effect on housing investment and significantly raise housing prices for a short period of time, both counter-intuitive to expectations. Phan (2014) uses four VAR models with differing strategies of identification to estimate the contribution of private consumption and investment to the response of output to monetary policy; each for Australia, US and UK over the period 1982-2007. The assessment of Australian, US and UK investment and consumption in response to monetary policy, facilitates a cross-country comparison of the composition of the output transmission mechanism of monetary policy. Phan estimates that Australian housing investment accounts for 35 per cent of the response of output, to monetary policy - far greater than the US or UK - and is cited as a major cause of the differences observed between Australian, US and UK responses to monetary policy shocks. The importance of the response of housing investment to the difference of the transmission of monetary policy is a conclusion that validates our assessment of housing investment as a separate channel of transmission within the housing market; in addition to the house price channel which is far more present in the existing literature. 3.4 Asset Price Targeting Literature: Literature detailing the effects of monetary policy upon housing output and prices has been discussed at length. However, this relationship has been observed in the reverse, whereby house prices inform monetary policy. This body of literature raises two questions, the first, whether central banks, should respond to asset price signals. The second, and most relevant to our research, is not a matter of whether central banks should respond โ€“ a normative question โ€“ but instead a positive question of whether monetary policy does respond to asset prices. The former is dealt with largely through theoretical literature, while whether central banks operate reactively to asset price signals is found in empirical literature. Literature discussing the validity of asset prices as informants of monetary policy and the view that monetary policy should react to asset price misalignments has developed rapidly in recent years. However, the idea that central banks need to account for asset prices in monetary policy is not a new phenomenon, first proposed by Irving Fisher who believed policy makers should try to stabilise a broad index of asset prices such as property, as well as the traditional consumer
  • 23. ย  ย  ย  15 goods and services index. Poole (1970) addressed potential benefits, discussing the โ€˜leaning against the windโ€™ monetary policy strategy that advocates changing interest rates when disturbances originate in asset markets. Kent and Lowe (1997) constructed a model that incorporated the notion of asset price misalignments into monetary policy, while Cecchetti et al (2002) argued that theoretical reasons exist for believing an inflation targeting central bank โ€“ such as the RBA โ€“ may improve macroeconomic performance by reacting to asset price misalignments. Zhao and Gao (2010), as a consequence of their Chinese analysis, recommended that asset prices be considered as endogenous variables in the monetary policy function, in order to make the objectives of monetary policy more controllable by the central bank. This conclusion is highly controversial and disagreed with by Bernanke and Fertler (2001), in part due to the difficulty in interpreting asset price misalignments. In addition to the debate over the merits of asset price targeting, papers have examined whether central banks already respond to asset prices. Zhao and Gao (2010) examined the impact of asset price fluctuation on Chinaโ€™s monetary policy, using quarterly data during the period of 1994-2006. The results from Granger causality and cointegration tests showed for every one per cent increase in housing prices, monetary policy interest rates increased by approximately two percent. Fiodendji (2012) examined whether Canadian monetary policy operates in response to asset prices. While the Canadian central bank admits consideration of asset prices in policy deliberations, due to their contribution to inflation, there has been explicit denial of an effort to directly stabilise asset prices. However, the findings suggest that stock market stabilisation plays a larger role than has been acknowledged. Additionally, Castro and Sousa (2010) found evidence of housing price accommodation in the monetary policy of both the US and UK. Whether reverse feedback exist between the Australian housing market and monetary policy, or rather whether the housing market โ€“ specifically house prices โ€“ influences monetary policy, will contribute to this growing literature. 3.5 Price Puzzle Literature: Forthcoming results from our research indicate the existence of a sectoral example of the price puzzle phenomenon, operating within the Australian housing market. We now move to briefly review the existing price puzzle literature, developing an understanding of how our research compares and contributes to the existing discussion. The term โ€˜price puzzleโ€™ was first coined
  • 24. ย  ย  ย  16 by Eichenbaum (1992) in his commentary on Simsโ€™ (1992) seminal paper studying the effects of monetary policy across countries, which identified in a SVAR model that a contractionary monetary policy shock โ€“ an increase in the federal funds rate โ€“ was associated with a counter- intuitive, persistent increase in the price level, ie the operation of a โ€˜price puzzleโ€™. Since Simsโ€™ discovery and Eichenbaumโ€™s formalisation of the price puzzle, a vast body of literature has developed. It has been reported that approximately 50 per cent of modern studies using VAR analysis to investigate the effects of monetary policy report that after a policy contraction prices increase, at least in the short run. (Rusnak et al 2013) One of the main explanations of the price puzzle is model mis-specification, particularly due to omitted variables, see (Rusnak et al 2013). Evidence of the price puzzle has been uncovered in both developed eg. US (Hanson 2004) and developing economies eg. Pakistan (Javid & Munir 2010), across the aggregated economy and within specific markets eg. the Chinese housing market (Yixun 2010) and the Australian housing market, as flagged in the findings of Wadud et al (2012) and Phan (2014). Further, research has been conducted into resolving the puzzle, focusing on how best to specify models to remove or reduce the puzzle. However, no formal, widely applicable approach has been developed and, no universal explanation for the puzzle exists. In sum, the price puzzle, remains as puzzling as ever. Our research contributes to the evidence of this anomalies existence. 3.6 Contribution to the Existing Literature: Our research fills a gap in the Australian literature where assessment of the role of the housing market in transmitting effects of monetary policy to inflation is sparse. While exhibiting similarity to Wadud et al (2012), in implementing a SVAR model to assess the response of the housing market to monetary policy, our analysis is differentiated by the intention to capture the effects of monetary policy on house prices and housing output, implementing a more focused SVAR that adopts fewer variables. Previous studies - particularly in the Australian context - have not directly assessed the importance the of housing market as a separate transmission mechanism of monetary policy, rather than a contributor to the interest rate and credit channels of transmission. Further differentiation comes from our focus in deciphering the role of each effect of monetary policy, on both house prices and housing investment, in maintaining
  • 25. ย  ย  ย  17 inflation. Lastly our consideration of possible reverse causation will serve to contribute to the growing debate over asset prices as monetary policy targets.
  • 26. ย  ย  ย  18 Chapter Four โ€“ Method and Data Following our review of the literature this chapter will outline the econometric method we follow in assessing the effects of monetary policy on house prices and housing output. This discussion involves explanation of ordinary least squares (OLS), the structural vector autoregressive (SVAR) model and the Cholesky identification process, among other econometric particulars. Furthermore, the data employed, in representation of several variables within the SVAR model, is outlined. 4.1 Ordinary Least Squares Method (OLS): OLS estimation of the relationship between monetary policy and the Australian housing market, facilitates preliminary analysis and provides an initial sense of the results. The relationship of house prices and output with monetary policy is estimated through the following pair of equations: OLS operates under the assumption of exogeneity of the independent variable ie. cash rate, and ignores endogeneity between variables. Endogeneity is fundamentally important to the reliability and accuracy of the relationships estimated, as well as the validity of statistical tests and also is relevant to our secondary concern of the housing market informing monetary policy. This necessitates the progression to a SVAR model, in which all variables are endogenous, so it is not reliant upon the assumption of exogeneity. ๐ป๐‘œ๐‘ข๐‘ ๐‘’ ย  ๐‘ƒ ๐‘Ÿ๐‘–๐‘๐‘’๐‘  ย  = ย  ๐›ผ + ๐›ฝ๐ถ๐‘Ž๐‘ โ„Ž ย  ๐‘… ๐‘Ž๐‘ก๐‘’ ย  + ๐‘’ ๐ป๐‘œ๐‘ข๐‘ ๐‘–๐‘›๐‘” ย  ๐‘‚ ๐‘ข๐‘ก๐‘๐‘ข๐‘ก = ย  ๐›ผ + ย  ๐›ฝ ๐ถ๐‘Ž๐‘ โ„Ž ย  ๐‘… ๐‘Ž๐‘ก๐‘’ + ๐‘’ (4.1) (4.2)
  • 27. ย  ย  ย  19 4.2 Structural Vector Autoregressive Model (SVAR): SVAR models are adopted in the overwhelming majority of research assessing monetary policy transmission. Sims (1980) introduced the use of VAR models for policy assessment. Prior to this formative paper the tool was used in forecasting, while simultaneous equation models were the mainstream method of economic analysis. The VAR model is used in economic analysis due to possible endogeneity of variables such as monetary policy โ€“ the cash rate โ€“ that simple OLS cannot account for. Thus, modelling is necessary for accurate and reliable analysis. Within the literature different forms of VAR are adopted, reduced form VAR, structural (SVAR) and cointegrated VAR models being the most common. The dominant use of the VAR model within the literature, to assess monetary policy transmission is for the most part uncontested, with the exception of the dynamic stochastic general equilibrium (DSGE) model. The DSGE model has been employed sparingly throughout the literature by the likes of Christidou & Konstantinou (2011) and Funke et al (2015). Although DSGE has been previously effective, the method is not ideally suited to empirical policy evaluation. In light of this the VAR model will be employed in our assessment of the importance of the Australian housing market to the transmission of monetary policy. The VAR model allows for dynamic feedback between variables and is useful in identifying effects from exogenous shocks such as monetary policy shocks. It is clear from the summary of existing literature which implemented VAR models, that the particular VAR form used is dependant on the nature of the data representing the variables of the model. In the case of our research, the specific VAR models considered are the SVAR and vector error correction (VEC) models. Whether the data sets exhibit evidence of non-stationarity and cointegration, determines the appropriateness of either the SVAR or VEC models. As the subsequent stationarity and autocorrelation results conclude, the estimation of a SVAR model is appropriate for our analysis. Our model takes the form of a three-variable SVAR system of the cash rate (CR), house prices (HP) and housing output (HQ) and may be expressed as per equation (4.3): ย 
  • 28. ย  ย  ย  20 ๐ถ๐‘…8 = ย  ๐›ฝ9 + ๐›ฝ:; < ;=9 ๐ถ๐‘…8>; + ย  ๐›ฝ?; < ;=9 ๐ป๐‘ƒ8>; + ย  ๐›ฝ@; < ;=9 ๐ป๐‘„8>; + ย  ๐œ–C8 ย  ๐ป๐‘ƒ8 = ย  ๐›พ9 + ๐›พ:; < ;=9 ๐ถ๐‘…8>; + ย  ๐›พ?; < ;=9 ๐ป๐‘ƒ8>; + ย  ๐›พ@; < ;=9 ๐ป๐‘„8>; + ย  ๐œ–E8 ย  ๐ป๐‘„8 = ย  ๐›ฟ9 + ๐›ฟ:; < ;=9 ๐ถ๐‘…8>; + ย  ๐›ฟ?; < ;=9 ๐ป๐‘ƒ8>; + ย  ๐›ฟ@; < ;=9 ๐ป๐‘„8>; + ย  ๐œ–G8 (4.3) 4.2.1 Identification of Monetary Policy Shock: While the merits of the VAR model are generally undisputed in the literature, the method for identification of the monetary policy shock is decidedly less unanimous. Ramey (2011) argued that in standard VAR shock-identification procedures, shocks are partially predictable and cannot be considered to be unexpected as assumed in theoretical analysis. Following Simsโ€™ (1980) seminal paper, and in line with Rameyโ€™s concern, many papers employ VAR models with different identification strategies. A common approach used to solve this problem, is the Cholesky decomposition, first proposed by Sims (1980). The Cholesky decomposition orthogonalises the shocks, including shocks to monetary policy, to recover the structural errors. However, with the Cholesky decomposition ordering of the variables is critically important; we account for this by repeating our model with alternative variable ordering in the sensitivity analysis of our results. 4.2.2 Impulse Response Functions: From our SVAR model we generate impulse response functions (IRF)s. An IRF graphically traces the temporal responses of each endogenous variable to an exogenous positive shock in an error, such as the one designed to capture monetary policy. We impose this shock as a one standard deviation increase in the monetary policy variable and include 2 standard errors and 95 per cent bootstrapped confidence intervals. IRFs allow graphical analysis of the direction, magnitude and persistence of the temporal responses to a monetary policy shock, thus enabling close examination of the relationships between monetary policy and the Australian housing market.
  • 29. ย  ย  ย  21 4.2.3 Block Exogeneity Wald Test: Upon estimation of our SVAR model and generation of IRFs, all variables are treated as endogenous. To formally assess the possibility of monetary policy acting in response to the Australian housing market we test the block exogeneity of the model, with respect to the cash rate; conducting a Wald test of block exogeneity which treats a previously endogenous variable - the cash rate - as exogenous and tests for joint significance of each of the other endogenous variables in the cash rate equation. The null hypothesis is the exogenous variable is only acting in response to lagged versions of itself. We specify this test algebraically in the forthcoming chapter, accompanied by discussion of its findings. Through this test we determine whether the cash rate, is in fact responding to the housing market or acts independently of these variables. 4.3 Variables: Having outlined the method this dissertation will follow, we now outline the variables implemented within the SVAR model. Existing literature has implemented an array of variables. However, the housing market model we derived in chapter two, relied only on the variables of house prices and housing output, in addition to the cash rate variable. Therefore, following the model of the housing market and in an effort to narrow the focus of our model we start from the minimum specifications necessary. As we cannot say anything sensible about the importance of the housing market to the transmission of monetary policy without including the three variables of most central relevance to the housing market model, the cash rate, house prices and housing output will be implemented, thus creating a three variable SVAR. We acknowledge different models will suggest the implementation of different additional variables; however, no one model may satisfy them all simultaneously, hence we reserve additional variables for implementation in sensitivity testing of the results of our three-variable SVAR model. These additional variables will include the Australian โ€“ Chinese exchange rate, growth of disposable income, commodity price index and a building cost variable.
  • 30. ย  ย  ย  22 4.4 Data: Our research is concerned with the Australian case of housing market monetary policy transmission and will therefore utilise Australian data. The data employed in our model spans the period 1990Q1 - 2015Q4, governed by the widespread consensus that operation of Australiaโ€™s central bank inflation targeting regime is observable from the early 1990s onward (Fraser 1993), while low inflation targets have been a feature of RBA rhetoric since as early as 1989 (Stevens 1999). For monetary policy, the RBA monthly cash rate target (CR) data is used, this data is available from 1990Q3 onward, again dictating our period of analysis. We extrapolate this data using the 90-day bank accepted bill rate for the two prior quarters to provide a rounded 25-year period. The use of CR is comparable to existing literature and is a particularly good fit for our research as the RBA uses the cash rate as its sole instrument, thus movement of the cash rate is fully illustrative of monetary policy. The Australian Bureau of Statistics (ABS) Consumer Price Index (CPI) is used to extract data representative of Australian house prices; as the purveyor of official housing market information, it is regarded as authoritative (Joye 2016). The data set includes an index of housing prices weighted across the eight capital cities: Adelaide, Brisbane, Canberra, Melbourne, Perth and Sydney and is recorded on a quarterly basis. Further, this data was also implemented in Wadud et al (2012) as representative of Australian house prices. The house price index (HPI) weights fluctuate throughout the 25-year period, yet new housing consistently accounts for over 70 percent; rental costs complete the index. The ABS Residential Property Price Index is not used as it only extends back to 2002. Housing output in our preferred model is represented by the ABS dwelling commencement (DC) data. This data records the number of new dwellings commenced each quarter within the private sector; both houses and other types of residential property ie. residential apartments. In our sensitivity analysis we make use of additional data sets. Firstly, an implicit price deflator (IPD) of housing prices is sourced from the ABS and used as an alternative to the CPI house price index. Further, gross fixed capital formation chain weighted index, with respect to new housing (GFCF) is used as an alternative to dwelling commencements, as is dwelling approvals (DA) which records the number of residential dwellings, both houses and other residential dwellings approved for construction. Data on housing finance commitments (FC) from the
  • 31. ย  ย  ย  23 ABS, capturing the value of finance both owner occupiers and investors accrue in purchasing new dwellings, is used as an additional proxy for housing output. Further, we examine the pass through of CR to interest rates, using the mortgage rate (MR) as alternative to the cash rate. MR data comes from the RBA; we use the standard variable owner occupier rate and believe it is appropriate due to Australiaโ€™s unique composition of mortgages, an overwhelming majority of which are variable rate mortgages. In addition to alternative data sets, we further our sensitivity analysis to reduce omitted variable bias, including a number of additional variables. We include a variable representative of the cost of building new homes (BC), using the ABS value of work data set to account for the increasing wages and housing materials. This is in line with the ABS approach to accounting for increasing costs of home building. Disposable income growth (YdG) data sourced from the ABS is included to account for increasing Australian disposable income. Further, we include data tracing the Australian to Chinese exchange rate (ACX), to examine the effect of Chinese real estate investment to house prices and housing outputs response to monetary policy. Lastly, in alignment with existing literature, we include a domestic commodity price index (CI) data from the RBA, in an effort to capture the forward looking rationale of monetary policy. All data is seasonally adjusted where available. Where appropriate, we convert data into its natural logarithm, transforming skewed distributions into more symmetrical distributions, and reducing scale discrepancies. We do not take the natural logarithm of CR or MR as they are rates. 4.4.1 Stationarity: The use of non-stationary data in SVAR models may produce unreliable results, as the standard error bounds of SVAR IRFs assume a stationary and normal distribution, while non-stationary data do not have the same distribution; ultimately leading to poor or incorrect understanding. Therefore, it is important to establish whether our data is stationary or non-stationary, and correct for non-stationary data. A preliminary manner of investigating stationarity is plotting each of the variables over time to observe whether the mean and variance appear dependent on time. In Figure 4.1 we graph CR, HPI and DC as a preliminary measure of helping fit our data.
  • 32. ย  ย  ย  24 Figure 4.1 Cash Rate, ln(House Price Index), ln(Dwelling Commencements) Plots Observable within the CR plot is the long period of monetary expansion that was enacted in the 1990s, subsequent to the announcement of the inflation target of two - three percent, and in response to the economic downturn of the early 1990s. This decline is observable until 1994. After six years of contraction, CR reduced in response to the GFC to what was then their lowest
  • 33. ย  ย  ย  25 level in history, three per cent. Subsequently, rates climbed again, peaking in 2011 at 4.75 per cent, then declining in an attempt to stimulate the economy which is yet to return to its pre GFC pace of activity, sitting at the close of 2015 at one and a half per cent. In examining the HPI plot we observe cyclical increases and decreases within the period 1990- 2000. However, this cyclical pattern is broken when HPI begins its strong, consistently upward trajectory. This unprecedented increase was not restricted to Australia only, but was rather, a worldwide phenomenon. The RBA has labelled the 2000s as when โ€œhousing put itself on the national agendaโ€ (Yates 2011); from this graph we can clearly see why the debate surrounding house prices and the house price bubble question remains as fiery as ever. The housing bubble problem was resolved for many economies after the GFC, as house prices dropped; however, this was not the case in Australia as HPI continues to steadily increase. DC shows a greater degree of volatility than HPI, yet the upward trend is similarly evident as DC displays an upward trend since 2012. The decreasing CR in the early 1990s are evident, as DC increased consistently. The effect of the introduction of the GST on the 1st July 2000 is also apparent, as housing investment was brought forward, causing a significant reduction in DC in the second half of 2000 (Kearns & Lowe 2011). The effects of the GFC are also apparent, accounting for the significant drop in 2008 that was subsequently reversed as a consequence of the contractionary monetary policy. In addition to revealing economic context, these plots are informative starting points in assessing stationarity. Upon initial examination HPI is expected to be non-stationary while DC may be stationary, CR could be stationary or perhaps a random walk with a negative drift. 4.5 Primary Variable Tests: To validate these ad hoc preliminary assessments, we move to a formal econometric stationarity assessment for which the augmented Dickey Fuller test (ADF) is used. The ADF test, pioneered by Dickey and Fuller (1981), is represented algebraically by equation (4.4):
  • 34. ย  ย  ย  26 โ–ณ ๐‘ฆ8 = ย  ๐›ผJ + ๐›ฝ๐‘ก + ย  ๐›พ ๐‘ฆ8>9 + ย  ๐›ฟฮ”๐‘Œ8>;M9 N ;=: + ย  ๐œ€8 The ADF test tests the null hypothesis H0: ฮณ = 0, meaning the time series has a present unit root, and is thus non-stationary. If the null hypothesis is rejected in favour of the alternative hypothesis HA: ฮณ โ‰  0, it is concluded that there is no unit root present and the time series is stationary. The ADF test is calculated across two deterministic specifications, intercept alone, assessing whether the series is a random walk around a drift or stationary, and an intercept and trend assessing whether the series is a random walk, or is stationary around an underlying deterministic trend. We calculate the ADF test at a four lag maximum, given the quarterly nature of our data. The p-values from the results are reported in Table 4.1; we accept or reject the null hypothesis at the five per cent significance level. Table 4.1 Unit Root Tests Level Data First Differences Lag Intercept Intercept + Trend No Intercept Intercept CRt 1 0.0163 0.0107 2 0.0086 0.0039 3 0.0032 0.0016 4 0.0165 0.0035 ln(HPIt) 1 0.9978 0.2621 0.0013 0.0049 2 0.9858 0.1421 0.0018 0.0059 3 0.9898 0.1070 0.0170 0.0489 4 0.9684 0.1446 0.0046 0.0087 ln(DCt) 1 0.0673 0.0475 2 0.0458 0.0374 3 0.0225 0.0127 4 0.2710 0.2631 The results of the ADF test on CR demonstrate that the null hypothesis should be rejected at all lags, across both deterministic specifications of the test; therefore, no unit root exists. CR is neither a random walk around a constant drift nor a random walk around a deterministic trend; the series is stationary. We fail to reject the null hypothesis when testing HPI, across all lags and both deterministic specifications, thus we conclude that in levels HPI is a non-stationary series. Moving to test HPI in first difference form, we can reject the null hypothesis, although not consistently across all lags and specifications. In first differences with an intercept, the null hypothesis is (4.4)
  • 35. ย  ย  ย  27 consistently rejected, thus with a constant drift, no unit root is present and HPI is stationary in first differences. When testing DC in levels under both deterministic specifications, we reject the null hypothesis inconsistently across the four lags. As the ADF test is predicated on the absence of autocorrelation, we move to test for the presence of autocorrelation in the residuals of the ADF equation at various lags. The Breusch-Godfrey serial correlation LM test is used to conduct this analysis. We test for autocorrelation of the first-fourth order as a consequence of our quarterly data. Under the same reasoning we continue to test to four lags. As reported by the p-values in Table 4.2, we may reject the null hypothesis of zero autocorrelation for lags 1 and 3 when an intercept is imposed and at lags 1 and 2 when a trend is also included; failing to reject the null hypothesis at the remaining lag specifications. In the ADF test at lag 2, with a constant and lag 3, with both a drift and deterministic trend we reject the null hypothesis of the presence of a unit root. Therefore, where we accept the Breusch-Godfrey null hypothesis of zero autocorrelation, and also reject the ADF null hypothesis of a present unit root, we conclude that DC is stationary.4 Table 4.2 Breusch-Godfrey Serial Correlation LM Test We have concluded that two variables, (CR and DC) are I(0), stationary and one (HPI) is I(1), non-stationary and therefore, cointegration is impossible. We correct for the I(1) variable by using the first differences of the series, estimating our primary SVAR model with CR, lnDC and the first difference of lnHPI. Estimating a SVAR model relies on the assumption of stationary time series data, we can now confirm that our primary model fulfills this assumption. 4 Since this conclusion is a relatively close call we will re-specify the model using the first differences of dwelling commencements, analysing the accumulated responses, comparing if the results are similar when using dwelling commencements in levels. Level Data Lag Intercept Intercept + Trend ln(DCt) 1 0.0074 0.0112 2 0.1044 0.0342 3 0.0423 0.0520 4 0.3415 0.2697
  • 36. ย  ย  ย  28 4.6 Robustness Variables Preliminary Statistics: 4.6.1 Alternative Data Sets In addition to the data of our preferred model, we employ several alternative data sets in our robustness testing. These are the mortgage rate (MR), implicit price deflator (IPD), dwelling approvals (DA), gross fixed capital formation index (GFCF) and finance commitments (FC). We now test the stationarity of these data using the same method followed previously. The ADF test is used to test the presence of a unit root in series, the p-values are reported in Table 4.3 and the Breusch-Godfrey test considers the existence of autocorrelation in the residuals, the p-values from which are reported in Table 4.4. Table 4.3 Unit Root Tests Level Data First Differences Lag Intercept Intercept + Trend No Intercept Intercept MRt 1 0.0057 0.0113 2 0.0004 0.0018 3 0.0100 0.0363 4 0.0133 0.0455 ln(IPDt) 1 0.9943 0.5624 0.0006 0.0000 2 0.9908 0.4688 0.0084 0.0004 3 0.9888 0.3924 0.0391 0.0062 4 0.9839 0.3509 0.0849 0.0212 ln(DAt) 1 0.0096 0.0083 0.0000 0.0000 2 0.1895 0.2279 0.0000 0.0000 3 0.1876 0.2299 0.0000 0.0002 4 0.0059 0.0066 0.0000 0.0000 ln(GFCFt) 1 0.4997 0.0026 0.0000 0.0000 2 0.6042 0.1091 0.0000 0.0001 3 0.4876 0.0476 0.0000 0.0000 4 0.6399 0.1514 0.0000 0.0001 ln(FCt) 1 0.3210 0.3113 0.0000 0.0000 2 0.3225 0.4739 0.0000 0.0000 3 0.1731 0.4849 0.0000 0.0000 4 0.1167 0.5189 0.0008 0.0002 Similarly, to CR, we reject the ADF null hypothesis for all lags and across deterministic specifications of MR, thus MR is stationary in its level form.
  • 37. ย  ย  ย  29 When testing IPD the null hypothesis of the ADF test cannot be rejected across all lag levels, regardless of whether a drift or trend is imposed. When we first difference the series we reject the null hypothesis inconsistently when no constant drift is imposed, yet when we impose a constant drift the null hypothesis of a present unit root can be consistently rejected. Therefore, we conclude when in first differences and a constant drift is imposed, the IPD series is stationary. We reject the ADF null hypothesis inconsistently across lags and deterministic specifications for DA, when in levels. We continue to test the DA series in levels under the Breusch-Godfrey test, testing the hypothesis of no first-fourth order residual autocorrelation. We reject for all lag levels except for the second, when a constant drift is imposed on the series. This lag and deterministic specification corresponds to a non-rejection of the ADF null hypothesis, therefore we conclude DA is non-stationary in levels. Moving to first differences, we can conclusively reject the ADF null hypothesis at all lags levels and deterministic specifications, therefore we conclude DA is stationary in first differences. The presence of a unit root in GFCF, cannot be rejected consistently across lags and deterministic specifications when the series is in levels. Examining autocorrelation in the residuals, we fail to reject the null hypothesis for all lags when a drift is imposed, except for the second, and cannot reject at any lag when a trend is also imposed. These lags correspond to ADF results that reject the null hypothesis of the ADF test, therefore we conclude GFCF is stationary in levels.5 When examining FC in levels we consistently fail to reject the ADF null hypothesis, therefore, in levels across lags and deterministic specifications the series is non-stationary. However, in first differences we consistently reject the ADF null hypothesis and conclude that FC is stationary, when in first differences regardless of whether a constant drift is imposed or not. 5 Again, we will look at the first differences in accumulates responses to confirm the validity of our conclusion since it is somewhat of a close call.
  • 38. ย  ย  ย  30 Table 4.4 Breusch-Godfrey Serial Correlation LM Test Level Data First Differences Lag Intercept Intercept + Trend No Intercept Intercept ln(IPDt) 1 0.4791 0.3656 0.0579 0.8395 2 0.8689 0.6531 0.2242 0.9192 3 0.8955 0.7938 0.6737 0.9661 4 0.9421 0.9650 0.8647 0.9471 ln(DAt) 1 0.0004 0.0004 0.0048 0.0056 2 0.0970 0.0433 0.0120 0.0136 3 0.0023 0.0023 0.0038 0.0038 4 0.0192 0.0145 0.0036 0.0038 ln(GFCFt) 1 0.0647 0.1483 0.3504 0.2278 2 0.0002 0.3160 0.4622 0.2766 3 0.2214 0.3823 0.9058 0.7210 4 0.4059 0.3984 0.6473 0.4112 4.6.2 Additional Variables In addition to the data in our preferred model, we employ several additional variables in our robustness testing of our primary results. These are disposable income growth (YdG), building costs (BC), Australian-Chinese exchange rate (ACX) and commodity price index (CI). We now test the stationarity of these data, the p-values from the ADF and Breusch-Godfrey tests are presented in Table 4.5 and 4.6, respectively. The ADF testing of YdG in level form, concludes that the null hypothesis is rejected across all lags and deterministic specifications. Therefore, YdG is stationary in levels. The ADF null hypothesis cannot be rejected consistently across deterministic specifications for BC when in levels. Looking to the Breusch-Godfrey test, we reject the null hypothesis of zero first-fourth order autocorrelation for all lags. When we transform BC into first differences, we consistently reject the ADF null hypothesis of non-stationarity across all lags and both specifications; concluding that BC in first differences is a stationary series. Both the ACX and CI series, when in levels consistently cannot reject the ADF null hypothesis, yet in first differences the null hypothesis is consistently rejected across all lags and both deterministic specifications. Therefore, in first differences, the ACX and CI series are stationary.
  • 39. ย  ย  ย  31 Table 4.5 Unit Root Tests Level Data First Differences Lag Intercept Intercept + Trend No Intercept Intercept ln(YdGt) 1 0.0000 0.0000 0.0000 0.0001 2 0.0000 0.0000 0.0000 0.0000 3 0.0001 0.0003 0.0000 0.0000 4 0.0070 0.0397 0.0000 0.0000 ln(BCt) 1 0.1852 0.0113 0.0000 0.0000 2 0.3123 0.0436 0.0000 0.0000 3 0.5165 0.2461 0.0000 0.0003 4 0.1171 0.0046 0.0000 0.0000 ln(ACXt) 1 0.0567 0.2979 0.0000 0.0000 2 0.0802 0.3639 0.0000 0.0000 3 0.0826 0.3660 0.0000 0.0000 4 0.1115 0.4477 0.0000 0.0000 ln(CIt) 1 0.5573 0.2585 0.0000 0.0000 2 0.6476 0.6054 0.0000 0.0000 3 0.7070 0.6214 0.0000 0.0000 4 0.7200 0.8788 0.0000 0.0000 Table 4.6 Breusch-Godfrey Serial Correlation LM Test Level Data First Differences Lag Intercept Intercept + Trend No Intercept Intercept ln(BCt) 1 0.0002 0.0000 0.0004 0.0005 2 0.0004 0.0000 0.0000 0.0000 3 0.0006 0.0003 0.0001 0.0001 4 0.0003 0.0007 0.0002 0.0001 In this chapter we looked at the base SVAR model, which we estimate in terms of CR, HPI and DC. We found HPI to be non-stationary, while DC and CR were both stationary. Therefore, we specify our base SVAR model in terms of CR, lnDC and the first difference of lnHPI. In addition, we specified the additional data and variables that will be used for sensitivity analysis. We found GFCF, MR and YdG to be stationary, whereas DA, IPD, FC, BC, AXC and CI are non-stationary and are specified in first differences to correct for this. Having determined our data is stationary, either in levels or first differences; we conclude the primary SVAR model and sensitivity analysis is fit for purpose. We examine our results in the forthcoming chapter.
  • 40. ย  ย  ย  32 Chapter Five - Results Having reviewed the existing literature, and the common empirical structural vector autoregression (SVAR) technique, we now combine the SVAR method and data described in the preceding chapter to analyse the responses of the Australian housing market to monetary policy. Our primary preoccupation is to examine the effects of monetary policy on house prices and housing output; effectively assessing the role of the Australian housing market in monetary policy transmission. Secondarily we seek to determine whether evidence exists to suggest monetary policy operates in response to the housing market. 5.1 Preliminary Analysis: 5.1.1 Ordinary Least Square Regression Analysis Elementary ordinary least squares (OLS) regression analysis provides us with an indication of the relationship between the cash rate and the housing market we may anticipate in our more sophisticated SVAR analysis. In Table 5.1 Panel A we report the results of two OLS regressions, equations (4.1) and (4.2); having individually regressed the house price index (HPI) and dwelling commencements (DC) on the cash rate (CR), from zero to two lags. HPI shows a positive relationship with CR from zero to one lag; at the second lag the relationship becomes negative and is our first significant result.6 Further, the magnitude of the negative relationship in the second lag is approximately equivalent to the cumulative magnitude of the positive relationships between the contemporaneous cash rate and the cash rate lagged one period. This suggests that HPI at first responds positively to CR; however, by the second lag this relationship becomes negative and 6 We continue to use the five per cent level of significance.
  • 41. ย  ย  ย  33 approximately outweighs the magnitude of the positive relations. This result is unexpected, suggesting HPI responds counter-intuitively at first to CR, yet when lagged by two periods this is corrected for, by a large and intuitive (negative) response. Table 5.1 OLS Regression Results Dependent Variable Independent Variable House Price Index (HPI) Dwelling Commencements (DC) Panel A Lag Coefficient P- Value Coefficient P -Value Cash Rate (CR) - 0.006638 (0.0780) 1.143227 (0.2557) Cash Rate (CR) 1 0.001578 (0.8063) -3.012966 (0.0033) Cash Rate (CR) 2 -0.008213 (0.0164) 3.968290 (0.0001) Panel B Cash Rate (CR) 1 0.011893 (0.0000) -0.078672 (0.0000) Cash Rate (CR) 2 -0.012221 (0.0000) 0.066659 (0.0000) Examining the results of this regression with respect to DC, it appears, that housing output responds intuitively (negatively) to CR quicker than house prices, as the result at lag 1, is not only significant, but negative and large in magnitude. Interestingly, at lag 2, this response appears to be outweighed by a larger positive response that is also significant. The unexpected relationship between HPI and CR could perhaps be the result of potential endogeneity issues and reverse causation with the contemporaneous CR. Thus, potentially causing the regression to capture two effects, not only that of CR to HPI, but also HPI to CR. We remove the contemporaneous CR to attempt to correct for this; the results are reported in Table 5.1 Panel B. When we remove the contemporaneous CR from the regressions we find that both HPI and DC coefficients become significant. The immediate and intuitive relationship between DC and CR continues to hold. The initial positive and counter-intuitive response of HPI reported in Panel A persists. However, now the negative response definitely outweighs the positive. The persistence of the initial counter-intuitive response of HPI, accompanied with the very small p-value suggests that the result is not simply due to endogeneity and reverse causation.
  • 42. ย  ย  ย  34 The OLS results presented are purely exploratory, as OLS regressions operate under the assumption of exogeneity of the independent variable (CR) and possible interaction between the housing prices and housing output is ignored. It is these constraints that motivate the use of the SVAR model, which relaxes the assumption of exogeneity and allows for dynamic interaction between variables. We now move to examine the results of our SVAR model analysis. 5.2 Structural Vector Auto-Regression Model: 5.2.1 SVAR (I) We will refer to our base model - shown algebraically in equation (4.3) - as SVAR(I), which models three primary variables; cash rate, house prices and housing output represented by the official cash rate (CR), house price index (HPI) and dwelling commencements (DC). 5.2.2 Lag Selection To ascertain the optimal lag level specification for SVAR(I) we compare the SVAR-lag- selection-criteria at a maximum of four lags.7 The criteria include the likelihood ratio (LR), final prediction error (FPE), Akaike information criterion (AIC), Shwartz information criterion (SIC) and Hannan-Quinn criterion (HQ). Reported in Table 5.2 are the results indicating which number of lags is recommended as optimal, according to each individual criterion. The results are mixed, some criteria recommending two lags, others four, although, the majority do in fact recommend to specify SVAR(I) at two lags. We consider serial correlation in the residuals of SVAR(I) at two and four lag orders, to resolutely conclude which lag specification is optimal. Table 5.2 ย  Optimal SVAR Lag Selection Criteria ย  Lag LogL LR FPE AIC SIC HQ 0 156.7792 NA 9.27e-06 -3.075585 -2.997430 -3.043954 1 394.5925 456.6016 9.54e-08 -7.651851 -7.339231 -7.525328 2 431.3878 68.43909 5.47e-08 -8.207755 -7.660669 -7.986340 3 436.3034 8.848142 5.95e-08 -8.126068 -7.344517 -7.809760 4 447.1006 18.78711 5.75e-08 -8.162012 -7.145995 -7.750812 7 Early experimentation ran up to 12 lags, however no important cases changed as a result of the increased maximum lag.
  • 43. ย  ย  ย  35 5.2.3 Residual Serial Correlation LM Test The null hypothesis of the residual serial correlation test is that no serial correlation is present in the residuals, shown algebraically below in equation (5.1), where ๐œŒ; represents residual autocorrelation at lag i. We test for first to fourth order autocorrelation, for both two and four lags, therefore i = two and four. Table 5.3 reports the results of our testing of first to fourth-order residual serial correlation at two and four lags. When specified with two lags, ie i=2, the null hypothesis is rejected at the fourth lag, suggesting the presence of fourth-order residual serial correlation; which is not entirely unexpected when using seasonally adjusted quarterly data. However, when tested with four lags, the fourth-order correlation is removed at the expense of the introduction of first- order serial correlation. Therefore, a four lag specification is not clearly superior to a two lag specification. We proceed to report SVAR(I) specified at two lags, in the interest of preserving simplicity in our model and reducing the degree of statistical noise introduced. ๐ปJ = ย  ๐œŒ; = ๐ถ๐‘œ๐‘ฃ ๐‘ˆ8 ย , ๐‘ˆ8>; ๐‘‰๐‘Ž๐‘Ÿ ๐‘ˆ8 = 0 ๐ปV = ๐œŒ; = ๐ถ๐‘œ๐‘ฃ(๐‘ˆ8 ย , ๐‘ˆ8>;) ๐‘‰๐‘Ž๐‘Ÿ(๐‘ˆ8) โ‰  0 (5.1)
  • 44. ย  ย  ย  36 Table 5.3 ย  SVAR Residual Serial Correlation LM Test ย  Tested at 2 lags Lags LM-Stat P-Value 1 11.05622 0.2719 2 5.730775 0.7665 3 15.75080 0.0723 4 20.96760 0.0128 Tested at 4 lags Lags LM-Stat P-Value 1 17.30800 0.0441 2 5.339288 0.8038 3 16.17685 0.0633 4 15.48913 0.0783 Probs from chi-square with 9 df. 5.2.4 Impulse Response Functions An impulse response function (IRF), graphically traces the responses of the three variables in SVAR(I) to an increase of each equationโ€™s standard error, which are orthogonal, based on the Cholesky decomposition. This allows us to observe the response of one variable to a shock or โ€˜impulseโ€™ in another, and thus study causality and the relationship shared between two variables in our model. Figure 5.1 presents all nine IRFs for SVAR(I), one for each variable (CR, HPI, DC) with respect to the one standard deviation shock to the error of each of the equations, over a 10 period time horizon. The IRFs of particular interest to our analysis are those in the bottom left corner that trace the response of HPI and DC to a shock in CR. Secondarily, are IRFs positioned in the top right hand corner, detailing the response of CR to HPI and DC shocks. The IRFs confirm what the OLS assessment signaled; HPI initially responds counterintuitively to a contractionary CR shock, increasing upon a CR increase. This behaviour is illustrative of the broader price puzzle phenomenon that has been documented across various markets, whereby
  • 45. ย  ย  ย  37 prices and thus inflation increase, in response to contractionary monetary policy. DC responded as theory would expect, decreasing significantly upon a positive CR shock, although more immediately, recalling our discussion of the housing market in chapter two, where we explained the response of housing output to changes in the housing market is understood to be delayed. We also learn from CR responses to shocks of HPI and DC that CR appears to be unresponsive to the Australian housing market: not shifting significantly upon either a HPI or DC shock. In the context of SVAR(I) this suggests that CR is exogenous to the Australian housing market and that monetary policy doesnโ€™t respond systematically to occurrences in the housing market. Less central to our analysis are the IRFs of the own effects of the variables, reported along the main diagonal, which as we would expect are all positive and significant. The interactions between HPI and DC are also as predicted, under the intuition of basic demand and supply of price and output variables: DC decreases upon a positive HPI shock while HPI increases upon a positive DC shock. When estimated at four lags, the results of SVAR(I) are comparable to the two lag specification. However, as expected, volatility - particularly the price puzzle behavior of HPI - is exacerbated.8 8 See Appendix Figure A.1, for IRFs of SVAR(I) over 10 periods, using a four lag specification.
  • 46. ย  ย  ย  38 Figure 5.1 Impulse Responses of SVAR(I) over 10 periods
  • 47. ย  ย  ย  39 5.2.5 Formal Test of Cash Rate Exogeneity Our OLS analysis and IRFs generated from our SVAR(I) model provide insight into whether CR is endogenous or in fact operates exogenously to the Australian housing market; both suggest that CR is exogenous. To formalise our assessment of whether monetary policy is exogenous within our SVAR(I) model, and thus independent of the Australian housing market, we execute the block exogeneity Wald test, shown algebraically below in equation (5.2): The null hypothesis of the test is that the coefficients of the excluded variables - in the case of SVAR(I), shown again in equation (5.3), these are house price index (HPI) and dwelling commencements (DC) - are jointly not significantly different to 0. The null hypothesis therefore concludes that the dependent variable - cash rate (CR) - does not operate in response to the excluded variables. In sum, the null hypothesis is that CR is exogenous to HPI and DC when considered jointly; and thus the Australian housing market. ๐ปJ = ย  ๐›ฝ?; = ย  ๐›ฝ@; = 0 ย  ๐ปV = ย  ๐›ฝ?; = ย  ๐›ฝ@; โ‰  0 ๐ถ๐‘…8 = ย  ๐›ฝ9 + ๐›ฝ:; < ;=9 ๐ถ๐‘…8>; + ย  ๐›ฝ?; < ;=9 ๐ป๐‘ƒ8>; + ย  ๐›ฝ@; < ;=9 ๐ป๐‘‚8>; + ย  ๐œ–C8 ย  ๐ป๐‘ƒ8 = ย  ๐›พ9 + ๐›พ:; < ;=9 ๐ถ๐‘…8>; + ย  ๐›พ?; < ;=9 ๐ป๐‘ƒ8>; + ย  ๐›พ@; < ;=9 ๐ป๐‘‚8>; + ย  ๐œ–E8 ย  ๐ป๐‘‚8 = ย  ๐›ฟ9 + ๐›ฟ:; < ;=9 ๐ถ๐‘…8>; + ย  ๐›ฟ?; < ;=9 ๐ป๐‘ƒ8>; + ย  ๐›ฟ@; < ;=9 ๐ป๐‘‚8>; + ย  ๐œ–G8 (5.3) (5.2)
  • 48. ย  ย  ย  40 Table 5.4 reports the p-value computed for the Wald Test. Based on the results, we cannot reject the null hypothesis and conclude in agreement with our OLS and IRF analysis, that CR is exogenous, and unresponsive to the Australian housing market, in the context of SVAR(I). By extension, we conclude in SVAR(I), that there exists no evidence of asset price targeting in Australian monetary policy, with respect to the Australian housing market. ย  Table 5.4 SVAR Block Exogeneity Wald Test Dependent Variable Cash Rate (CR) Excluded Variables P-Value ln(HPIt), ln(DCt) 0.5709 5.3 Sensitivity Analysis: Upon specification of our base model, SVAR(I) a multitude of decisions, many of which we do not have firm theoretical ground to guide our choices, were made. Choices regarding the inclusion of variables, and use of data are examples of decisions that another reasonable researcher may have specified differently. In our sensitivity analysis we assess the robustness of the three principal findings of SVAR(I): the price puzzle behavior of HPI in response to CR, the intuitive response of DC to CR, and the exogeneity of the CR to the Australian housing market by replacing our base model specifications with sensible alternative assumptions. We will now outline the sensitivity analysis carried out on the findings of SVAR(I), and draw attention only to cases where the findings were no longer robust or are particularly interesting. All sensitivity analysis is reported at the two lag specification, for cases where a lag specification other than two was optimal we will draw attention to this.
  • 49. ย  ย  ย  41 5.3.1 ย  Alternative Variable Representation A. ย  Implicit Price Deflator (IPD) substituted for HPI B. ย  Gross Fixed Capital Formation (GFCF) substituted for DC C. ย  Dwelling Approvals (DA) substituted for DC D. ย  Finance Commitments (FC) substituted for DC We substitute alternate data sets as proxies for house prices and housing output. We make use of the following data sets, implicit price deflator (IPD) as representative of house prices, dwelling approvals (DA), gross fixed capital formation index (GFCF) and finance commitments (FC), as representative of housing output. Results of (A), (B) and (C) substantiate the three central findings of SVAR(I).9 The price puzzle persists and the response of housing output is comparable to SVAR(I) when IPD is substituted for HPI and again when DA and GFCF are substituted for DC, although the volatility of both is exacerbated. Further, CR remains unresponsive to the housing market, as the IRFs show minimal movement, and more formally as we fail to reject the null hypothesis of the block exoegeneity test.10 Finance Commitments The results of (D) - where FC are substituted for DC as representative of housing output - substantiate the first and second main findings of SVAR(I), that of the price puzzle behavior of house prices and the intuitive response of housing output, although the latter appears differently to SVAR(I).11 Although from first glance at Figure 5.2, the timing of the response of FC to a CR shock does not appear comparable to SVAR(I), a similar story is in fact being told. FC decreases immediately to a positive CR shock, and begins to return to its original level immediately after that. When we consider FC as the proxy for housing output, we understand that FC precedes DA, DC and housing investment, explaining the timing differences between the response of housing output present in D and SVAR(I). 9 The IRFs are reported in the Appendix as Figure A.2, A.3 and A.4, respectively. 10 Results of the block exogeneity Wald test are reported in the Appendix Table A.1. 11 FC measures the numerical value of finance undertaken by those purchasing new homes, both owner occupiers and investors.
  • 50. ย  ย  ย  42 The finding of CR exogeneity, however, no longer holds as it is observable in Figure 5.2 that CR is responsive to a FC shock. This is confirmed by the p-values of the block exogeneity test - although only just significant - reported in Table 5.5. We reject the null hypothesis of the test, concluding that under the alternative specifications of (D), CR is responsive to the Australian housing market. Further, in the Appendix Table A.2, we consider Granger causality and conclude FC is the Granger cause of CR; we do not find HPI to be the Granger cause of CR. Interestingly this result is inconsistent to the finding of CR exogeneity in SVAR(I). This difference potentially suggests that the RBA is responsive to the extent to which people accumulate debt in the housing market, irrespective of the level of DC or housing output, per se. Further, as FC occurs before DC, perhaps the RBA reacts to FC, as a predictor of future housing activity, even though it does not react to DC. Figure 5.2 Impulse Responses of FC and CR over 10 Periods ย 
  • 51. ย  ย  ย  43 Table 5.5 SVAR Block Exogeneity Wald Test 5.3.2 ย  Additional Variables In their research into the price puzzle in VAR analysis, Rusnak et al (2011) argue its presence is largely due to model mis-specification, rather than an actual case of the price puzzle. The paper argues that most cases of the price puzzle can be eliminated or significantly reduced when the model is correctly specified, particularly when omitted variables are included. In an effort to reduce omitted variable bias we introduce a range of variables potentially significant to our results: A. ย  Building Costs (BC) B. ย  Disposable Income Growth (YdG) C. ย  Australian โ€“ Chinese Currency Exchange Rate (ACX) D. ย  Commodity Price Index (CI) BC is included to capture the increasing costs of homebuilding, YdG is included to account for the possibility that disposable income growth may be driving house price increases. Chinese demand for Australian property is by far the largest foreign demand in the Australian housing market, as the Foreign Investment Review Board reported Chinese purchasers spent $24.3 billion on Australian property in 2014-2015 alone (Maley 2016). Anecdotally it has been suggested that this strong Chinese demand for Australian property is driving house price increases. We include the Australian โ€“ Chinese exchange rate (ACX) as a variable intended to tease out the effects of Chinese demand. Lastly, in previous literature, including the paper by Rusnak, Havranek and Horvath (2011) it has been raised that the inclusion of commodity prices (CI) may alleviate price puzzle behavior from a SVAR model that was mis-specified. The inclusion of CI stems from Simโ€™s (1992) theory that central banks respond to expectations of higher future inflation by contracting the cash rate, but not enough to prevent inflation from Dependent Variable Cash Rate Excluded Variables P-Value ln(HPIt) ln(FCt) 0.0409
  • 52. ย  ย  ย  44 actually rising, and that the central bank has information about future inflation that is not captured in the SVAR model. It is understood that CI captures this information. Initially we introduce these additional variables one at a time at the end of our model, ordered last, after DC. We recognise the ordering of variables when using a Cholesky decomposition can be important, and will revisit the order later in our analysis. Results from (A), (B), (C) and (D) all corroborate the price puzzle behavior of HPI and the intuitive response of DC to a positive CR shock.12 CR exogeneity is robust for only (A) and (B); in these cases we fail to reject the null hypothesis of the block exogeneity when estimated at 2 lags.13 We will deal with the exceptions in more detail. Exogeneity does not hold when we estimate (D) at two lags.14 Upon assessment of Granger causality, we find CI to be a Granger cause of CR; HPI and DC are not Granger causes of CR.15 However, when we estimate (C) at its optimal four lag specification, CR exogeneity holds. We will now focus specifically on the case of (C) and return to (D) when considering variable ordering. Australian โ€“ Chinese Exchange Rate In Figure 5.3 we observe evidence of CR responding to a positive ACX shock; appreciation of the Australian dollar against the Chinese Yuan. From the p-value reported in Table 5.6 we reject the null hypothesis of the block exogeneity test for the specification of (C), therefore, the finding of CR exoegeneity is no longer robust, and CR is deemed endogeneous to theAustralian housing market. Further, as reported in the Appendix Table A.5, ACX is a Granger cause of CR. HPI and DC are found to continue to not be Granger causes of CR. CR continues to be unresponsive to HPI and DC (see Appendix Figure A.8), yet responsive to ACX, suggesting that the RBA is influenced more by exchange rates in policy than the housing market. This is quite plausible given the public attention former Governor Glenn Stevens drew to exchange rates during his time at the RBA. 12 See Appendix Figures A.5 โ€“ A.10. Where 4 lags were optimal (A & D) we present at both two and four lags. 13 See Appendix Table A.3. 14 See Appendix Table A.4. 15 See Appendix Table A.5 ย 
  • 53. ย  ย  ย  45 Table 5.6 SVAR Block Exogeneity Wald Test Dependent Variable Cash Rate Excluded Variables P-Value ln(HPIt), ln(DCt), ln(ACXt) 0.0421 Variable Ordering In using Cholesky decomposition, the ordering of variables in our model is potentially significant to the results our model yields. We do not alter the ordering of our base variables (CR, HPI, DC), as the original order, in terms of responsiveness makes intuitive sense. We reorder the additional variables (A) - (D), from first in our model to third. The reordering of our additional variables does not alter their effects, the results from these specifications continue to reinforce the findings of SVAR(I), with the exclusion of (C), as CR continues to be responsive to ACX and the housing market. The reordering of perhaps the most interest and potential significance is the ordering of CI first in our model, before the cash rate, as this specification is in alignment with existing literature that argues the inclusion of CI is effective in reducing - if not removing - evidence of a price puzzle, as CI, when ordered first in the SVAR captures the future inflation expectations the central bank is believed to be Figure 5.3 Impulse Response of CR to ACX over 10 Periods ย